Authorship：Last author,　Corresponding author   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/ICASSP48485.2024.10446402

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icassp/icassp2024.html#DuWWH24

Authorship：Lead author,　Last author,　Corresponding author   Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.neucom.2024.127500

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.compbiolchem.2024.108166

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.bspc.2023.105748

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/JBHI.2023.3305644

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1587/transinf.2022edp7193

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1007/s00371-022-02452-y

researchmap

Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/PerComWorkshops56833.2023.10150371

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/percom/percomw2023.html#InoshitaNNIIMH23

Authorship：Last author,　Corresponding author   Publishing type：Research paper (international conference proceedings)  

DOI： 10.23919/MVA57639.2023.10215720

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mva/mva2023.html#TadaH23

Authorship：Last author   Publishing type：Research paper (international conference proceedings)  

DOI： 10.23919/MVA57639.2023.10215580

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mva/mva2023.html#MoritaH23

Authorship：Last author   Publishing type：Research paper (international conference proceedings)  

DOI： 10.23919/MVA57639.2023.10216087

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mva/mva2023.html#HaradaH23

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.3390/info14010033

researchmap

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1145/3551626.3564946

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mmasia/mmasia2022.html#NambaH22

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1145/3551626.3564958

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mmasia/mmasia2022.html#YamawakiH22

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/MIPR54900.2022.00040

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mipr/mipr2022.html#YamawakiH22

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1007/978-3-031-16452-1_21

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/miccai/miccai2022-8.html#LiIHLHC22

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.24963/ijcai.2022/135

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/ijcai/ijcai2022.html#HuangXLIH0022

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/ICIP46576.2022.9897182

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icip/icip2022.html#0039H22

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/ICIP46576.2022.9897424

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icip/icip2022.html#0039H0022

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/ICIP46576.2022.9897926

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icip/icip2022.html#YamawakiKH22

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/GCCE56475.2022.10014204

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/gcce/gcce2022.html#WangWYLLHLTHC22

Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/EMBC48229.2022.9871672

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/embc/embc2022.html#YangICXCHHTL22

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/EMBC48229.2022.9871188

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/embc/embc2022.html#AnandaIHLHC22

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/EMBC48229.2022.9871539

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/embc/embc2022.html#JainSWNIHLHRC22

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/EMBC48229.2022.9871896

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/embc/embc2022.html#AiAWLIHC22

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/CVPRW56347.2022.00072

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/cvpr/cvpr2022w.html#NanbaMH22

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1109/JBHI.2022.3153406

researchmap

Authorship：Last author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ISBI52829.2022.9761578

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/isbi/isbi2022.html#NakaiH22

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICCE53296.2022.9730129

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/iccel/icce2022.html#LiIHLTHFKC22

Language：English   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/ICASSP43922.2022.9746172

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icassp/icassp2022.html#WangXLIHCT22

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (scientific journal)  

DOI： 10.1145/3477396

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/TCSVT.2021.3112551

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：MDPI  

The goal of single image super resolution (SISR) is to recover a high-resolution (HR) image from a low-resolution (LR) image. Deep learning based methods have recently made a remarkable performance gain in terms of both the effectiveness and efficiency for SISR. Most existing methods have to be trained based on large-scale synthetic paired data in a fully supervised manner. With the available HR natural images, the corresponding LR images are usually synthesized with a simple fixed degradation operation, such as bicubic down-sampling. Then, the learned deep models with these training data usually face difficulty to be generalized to real scenarios with unknown and complicated degradation operations. This study exploits a novel blind image super-resolution framework using a deep unsupervised learning network. The proposed method can simultaneously predict the underlying HR image and its specific degradation operation from the observed LR image only without any prior knowledge. The experimental results on three benchmark datasets validate that our proposed method achieves a promising performance under the unknown degradation models.

DOI： 10.3390/electronics10212591

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

Precise delineation of target tumor from positron emission tomography-computed tomography (PET-CT) is a key step in clinical practice and radiation therapy. PET-CT co-segmentation actually uses the complementary information of two modalities to reduce the uncertainty of single-modal segmentation, so as to obtain more accurate segmentation results. At present, the PET-CT segmentation methods based on fully convolutional neural network (FCN) mainly adopt image fusion and feature fusion. The current fusion strategies do not consider the uncertainty of multi-modal segmentation and complex feature fusion consumes more computing resources, especially when dealing with 3D volumes. In this work, we analyze the PET-CT co-segmentation from the perspective of uncertainty, and propose evidence fusion network (EFNet). The network respectively outputs PET result and CT result containing uncertainty by proposed evidence loss, which are used as PET evidence and CT evidence. Then we use evidence fusion to reduce uncertainty of single-modal evidence. The final segmentation result is obtained based on evidence fusion of PET evidence and CT evidence. EFNet uses the basic 3D U-Net as backbone and only uses simple unidirectional feature fusion. In addition, EFNet can separately train and predict PET evidence and CT evidence, without the need for parallel training of two branch networks. We do experiments on the soft-tissue-sarcomas and lymphoma datasets. Compared with 3D U-Net, our proposed method improves the Dice by 8% and 5% respectively. Compared with the complex feature fusion method, our proposed method improves the Dice by 7% and 2% respectively. Our results show that in PET-CT segmentation methods based on FCN, by outputting uncertainty evidence and evidence fusion, the network can be simplified and the segmentation results can be improved.

DOI： 10.1088/1361-6560/ac299a

PubMed

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Science and Business Media Deutschland GmbH  

Isocitrate dehydrogenase (IDH) status is an important basis for the diagnosis of gliomas in the 2016 World Health Organization classification scheme. A strong relationship exists between IDH mutation status and glioma prognosis. The preoperative prediction of IDH status is essential for the treatment of gliomas. However, the existing medical methods cannot predict IDH status before an operation. In this study, we propose a modality self-attention network to predict IDH mutation status on multimodality magnetic resonance imaging images. The proposed method predicts the importance of each modality for classification task and calculates weights and then uses weighted images for training. Moreover, we select a light and high-performance self-attention network for the classification to solve the overfitting problem on the glioma dataset of the First Affiliated Hospital of Zhengzhou University (FHZU). The proposed method achieved an F1-score of 0.6570 on the FHZU dataset, which is better than SE-Net (0.2563), a method proposed by Yoon Seong Choi et al. (0.3999), and SA-Net (0.5245).

DOI： 10.1007/978-981-16-3013-2_5

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Science and Business Media Deutschland GmbH  

Deep Learning provides exciting solutions to problems in medical image analysis and is regarded as a key method for future applications. However, only a few annotated medical image datasets exist compared to numerous natural images. A solution to this problem is transfer learning using ImageNet. However, because the domain of ImageNet is different from that of medical images, the results of transfer learning are not always good. Therefore, we propose a model to investigate transfer learning by self-supervised learning using medical images. It is widely known that the results of Computerized Tomography (CT) scan are 3D volume images. There are lots of slices in CT or Magnetic Resonance Imaging scan images. So why not make these slices to a class? It is imperative to formulate this intuition as a self-supervised feature learning at the case-level. The results of our experiment demonstrated that, under self-supervised feature learning settings, our method surpasses the transfer learning method that uses ImageNet for classification. By experimenting with unannotated datasets, our method is remarkable for consistently improving test performance with a few annotated data. By fine-tuning the learned features, we obtained competitive results for self-supervised learning and classification tasks.

DOI： 10.1007/978-981-16-3013-2_20

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Science and Business Media Deutschland GmbH  

Content based image retrieval (CBIR) that searches for similar images in a large database has been attracting increasing research interest recently, and it has been applied to medical image characterization for sharing experts’ experiences. One challenging task in CBIR is to extract features for effective image representation. To this end, bag-of-visual-words (BoVW) has been proven to be effective to extract middle-level features for image analysis. However, it is necessary to first vectorize the two- or three-dimensional spatial structure for analysis in conventional BoVW and then destroy the spatial relationships of nearby voxels. In this study, we propose a tensor sparse coding method, which is a multilinear generalization of conventional sparse coding (soft assignment in BoVW), to learn features from multi-dimensional medical images. We regard high-dimensional local structures as tensors and propose a K-CP (CANDECOMP/PARAFAC) algorithm to learn an overcomplete tensor dictionary iteratively. By using the learned overcomplete tensor dictionary, sparse coefficients of tensor local structures are calculated by employing the tensor orthogonal matching pursuit (Tensor-OMP) algorithm, which is an extended multilinear version of the conventional vector-based OMP. The proposed method is applied to the retrieval of focal liver lesions (FLLs) by using a medical database consisting of contrast-enhanced multi-phase computer-tomography (CT) images. Experiments show that the proposed tensor sparse coding method achieved better retrieval performance than conventional methods.

DOI： 10.1007/978-981-16-3013-2_18

Scopus

researchmap

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/MIPR51284.2021.00036

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mipr/mipr2021.html#FangHQJC21

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-030-87196-3_39

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/miccai/miccai2021-2.html#HuangZLHIHCT21

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/ICIP42928.2021.9506783

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icip/icip2021.html#YamawakiH21

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/ICCVW54120.2021.00138

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/iccvw/iccvw2021.html#KoheiH21

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICASSP39728.2021.9413957

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icassp/icassp2021.html#HuangCLZMQP0IH021

Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/EMBC46164.2021.9630703

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/embc/embc2021.html#AonpongIHLC21

Publishing type：Research paper (international conference proceedings)  

DOI： 10.1109/EMBC46164.2021.9629671

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/embc/embc2021.html#LiIHFKC21

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/TMI.2021.3089661

researchmap

Authorship：Corresponding author   Publishing type：Research paper (scientific journal)  

DOI： 10.1109/ACCESS.2021.3122450

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/ACCESS.2021.3088234

researchmap

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.23919/MVA51890.2021.9511356

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mva/mva2021.html#NakaiQH21

Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

DOI： 10.1145/3460426.3463673

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/mir/icmr2021.html#WangHLH021

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-030-87193-2_13

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/miccai/miccai2021-1.html#WangLHCLIHCT21

Publishing type：Research paper (scientific journal)  

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/corr/corr2104.html#abs-2104-14420

Publishing type：Research paper (scientific journal)  

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/corr/corr2103.html#abs-2103-04235

Publishing type：Research paper (scientific journal)  

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/corr/corr2103.html#abs-2103-00274

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICCE50685.2021.9427760

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/iccel/icce2021.html#HasegawaIHLHC021

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/TCBB.2019.2919015

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/tcbb/tcbb18.html#Kitrungrotsakul21

Authorship：Corresponding author   Publishing type：Research paper (scientific journal)  

DOI： 10.3390/s21072348

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

Automated nuclei recognition and detection is a critical step for a number of computer assisted pathology based on image processing techniques. However, automated nuclei recognition and detection is quite challenging due to the exited heterogeneous characteristics of cancer nuclei such as large variability in size, shape, appearance, and texture of the different nuclei. Deep learning approaches, where the most popular one is the deep Convolutional Neural Network (CNN), have been shown to provide encouraging results in different computer vision tasks, and many CNN models learned already with large-scale image dataset such as ImageNet have been released. How to effectively adopt the exiting CNN models to other domain tasks such as medical image analysis has attracted hot attention for transferring the obtained knowledge from the general image set to the specific domain task, which is called as transfer learning. Since the released CNN model usually require a fixed size of input images, transfer learning strategy compulsorily unifies the available images in the target domain to the required size in the CNN models, which maybe modifies the inherent structure in the target images and affect the final performance. This study exploits an adaptable transfer learning strategy flexibly for any size of input images via removing the mathematical operation components but retaining the learned knowledge in the exiting CNN models. We modify the released CNN models: AlexNet, VGGnet and ResNet previously learned with the ImageNet dataset for dealing with the small-size of image patches to implement nuclei recognition. Experimental results show that our proposed adaptable transfer learning strategy achieves promising performance for nuclei recognition compared with a constructed CNN architecture for small-size of images.

DOI： 10.1007/978-981-15-3250-4_16

Scopus

researchmap

Language：English   Publishing type：Part of collection (book)   Publisher：Springer  

Image classification is to assign one or more labels to an image, which is one of the most fundamental tasks in computer vision and pattern recognition. In traditional image classification, low-level or mid-level features are extracted to represent the image and a trainable classifier is then used for label assignments. In recent years, the high-level feature representation of deep convolutional neural networks has proven to be superior to hand-crafted low-level and mid-level features. In the deep convolutional neural network, both feature extraction and classification networks are combined together and trained end-to-end. Deep learning techniques have also been applied to medical image classification and computer-aided diagnosis. In this chapter, we first introduce fundamentals of deep convolutional neural networks for image classification and then introduce an application of deep learning to classification of focal liver lesions on multi-phase CT images. The main challenge in deep-learning-based medical image classification is the lack of annotated training samples. We demonstrate that fine-tuning can significantly improve the accuracy of liver lesion classification, especially for small training samples.

DOI： 10.1007/978-3-030-32606-7_3

Scopus

researchmap

Language：English   Publishing type：Part of collection (book)   Publisher：Springer  

In this work, we aim at classification and quantification of emphysema in computed tomography (CT) images of lungs. Most previous works are limited to extracting low-level features or mid-level features without enough high-level information. Moreover, these approaches do not take the characteristics (scales) of different emphysema into account, which are crucial for feature extraction. In contrast to previous works, we propose a novel deep learning method based on multi-scale deep convolutional neural networks. There are three contributions for this paper. First, we propose to use a base residual network with 20 layers to extract more high-level information. Second, we incorporate multi-scale information into our deep neural networks so as to take full consideration of the characteristics of different emphysema. A 92.68% classification accuracy is achieved on our original dataset. Finally, based on the classification results, we also perform the quantitative analysis of emphysema in 50 subjects by correlating the quantitative results (the area percentage of each class) with pulmonary functions. We show that centrilobular emphysema (CLE) and panlobular emphysema (PLE) have strong correlation with the pulmonary functions and the sum of CLE and PLE can be used as a new and accurate measure of emphysema severity instead of the conventional measure (sum of all subtypes of emphysema). The correlations between the new measure and various pulmonary functions are up to |r| (r is correlation coefficient).

DOI： 10.1007/978-3-030-32606-7_9

Scopus

researchmap

Language：English   Publishing type：Part of collection (book)   Publisher：Springer  

Self-taught learning aims at obtaining compact and latent representations from data them-selves without previously manual labeling, which would be time-consuming and laborious. This study proposes a novel self-taught learning for more accurately reconstructing the raw data based on the sparse autoencoder. It is well known that autoencoder is able to learn latent features via setting the target values to be equal to the input data, and can be stacked for pursuing high-level feature learning. Motivated by the natural sparsity of data representation, sparsity has been imposed on the hidden layer responses of autoencoder for more effective feature learning. Although the conventional autoencoder-based feature learning aims at obtaining the latent representation via minimizing the reconstruction error of the input data, it is unavoidable to produce reconstruction residual error of the input data and thus some tiny structures are unable to be represented, which may be essential information for fine-grained image task such as medical image analysis. Even with the multiple-layer stacking for high-level feature pursuing in autoencoder-based learning strategy, the lost tiny structure in the former layers cannot be recovered evermore. Therefore, this study proposes a residual sparse autoencoder for learning the latent feature representation of more tiny structures in the raw input data. With the unavoidably generated reconstruction residual error, we exploit another sparse autoencoder to pursuing the latent feature of the residual tiny structures and this self-taught learning process can continue until the representation residual error is enough small. We evaluate the proposed residual sparse autoencoding for self-taught learning the latent representations of HEp-2 cell image, and prove that promising performance for staining pattern recognition can be achieved compared with the conventional sparse autoencoder and the-state-of-the-art methods.

DOI： 10.1007/978-3-030-32606-7_11

Scopus

researchmap

Publishing type：Research paper (scientific journal)  

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/corr/corr2010.html#abs-2010-09241

Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-030-59719-1_42

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/miccai/miccai2020-4.html#MoCLTCWHIHC20

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICPR48806.2021.9412321

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icpr/icpr2020.html#KoheiH20

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-030-69532-3_5

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/accv/accv2020-2.html#YamamichiH20

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-030-69756-3_5

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/accv/accv2020w.html#FujitaH20

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-030-69756-3_3

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/accv/accv2020w.html#LiuZH20

Authorship：Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

DOI： 10.1007/978-3-030-69756-3_1

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/accv/accv2020w.html#FangH20

Publishing type：Research paper (scientific journal)  

DOI： 10.4310/cis.2020.v20.n4.a2

researchmap

Publishing type：Research paper (scientific journal)  

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/corr/corr2004.html#abs-2004-08790

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ISBI45749.2020.9098654

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/isbi/isbi2020.html#HuangLTHZIHCW20

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICCE46568.2020.9043172

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/iccel/icce2020.html#YoshinobuIHLHZC20

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICASSP40776.2020.9053405

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icassp/icassp2020.html#HuangLTHZIHCW20

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (scientific journal)  

DOI： 10.1145/3386313

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/JBHI.2019.2963195

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1007/s00521-019-04374-8

researchmap

Authorship：Corresponding author   Publishing type：Research paper (scientific journal)  

DOI： 10.1007/s11042-018-6751-5

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.patrec.2019.01.001

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

Convolutional neural networks (CNNs) have achieved great success in numerous challenging vision tasks, and have great potential for object detection in natural images. Compared with the natural images, medical images exhibit some unique characteristics. Therefore, substantial challenges still remain in this field. The first challenge is to develop a method for effectively distilling enhancement patterns from the dynamic CT images. Moreover, since tumor sizes vary greatly and small lesions are important for early liver tumor detection, lesion detection with a widely variable scale is another challenge. In this paper, we propose a multi-stream scale-insensitive convolutional and recurrent neural network (MSCR) for liver tumor detection. Specifically, we propose the use of grouped convolutional long short-term memory (GCLSTM) to extract enhancement patterns, which is developed as a plug-and-play module. Experiments show that the MSCR framework exhibits superior performance over state-of-the-art approaches, achieving an average precision of 77.06% for detection of focal liver lesions. We have released the code of MSCR in 1

DOI： 10.1109/ICIP.2019.8803730

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

Automatic liver lesion classification on computed tomography images is of great importance to early cancer diagnosis and remains a challenging task. State-of-the-art liver lesion classification algorithms are currently based on manually selected regions of interest (ROIs) or automatically detected ROIs. However, liver lesions usually vary in size and shape, which makes the ROI selection process labor-intensive and also poses an obstacle to automatic lesion detection. In this paper, we propose a dual-attention dilated residual network (DADRN) as a potential solution to lesion classification task without manual ROI selection or automatic lesion detection. We incorporated a novel dual-attention module in order to capture the non-local feature dependencies and help the deep neural network focus on the lesion area by enlarging the difference between the lesion area and nonlesion area. To the best of our knowledge, we are the first to employ the self-attention mechanism to address liver lesion classification task. In addition, the well-trained DADRN can be used for weakly-supervised lesion localization without any architectural change or retraining. Experiment results show that DADRN could achieve a lesion classification accuracy comparable to that of the state-of-the-art ROI-based method and outperformed state-of-the-art attention-based approaches in both liver lesion classification and localization tasks.

DOI： 10.1109/ICIP.2019.8803009

Scopus

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-030-32226-7_17

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-030-33391-1_15

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/EMBC.2019.8856356

researchmap

Other Link： https://dblp.uni-trier.de/conf/embc/2019

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/JBHI.2018.2890045

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/titb/titb23.html#PengCLHLCLWHI19

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICCVW.2019.00533

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/iccvw/iccvw2019.html#HanZC19

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/BigMM.2019.00028

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/bigmm/bigmm2019.html#HanSC19

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/BigMM.2019.00-13

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/bigmm/bigmm2019.html#HanC19

Authorship：Lead author,　Corresponding author   Publishing type：Research paper (scientific journal)  

DOI： 10.3390/s19245401

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/sensors/sensors19.html#HanSWSZC19

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In this paper, we present an automatic approach to paranasal sinus segmentation in computed tomography (CT) images. The proposed method combines a probabilistic atlas and a fully convolutional network (FCN). The probabilistic atlas was used to automatically localize the paranasal sinus and determine its bounding box. The FCN was then used to automatically segment the paranasal sinus in the bounding box. Comparing our proposed method with the conventional FCN (without probabilistic atlas) and the state-of-the-art method using active contour with group similarity, the proposed method demonstrated an improvement in the paranasal sinus segmentation. The segmentation accuracy (Dice coefficient) was about 0.83 even for the case with unclear boundary.

DOI： 10.1109/EMBC.2019.8856703

PubMed

researchmap

Other Link： https://dblp.uni-trier.de/conf/embc/2019

Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.compmedimag.2019.05.002

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/cmig/cmig75.html#Kitrungrotsakul19

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/ICASSP.2019.8682326

researchmap

Other Link： https://dblp.uni-trier.de/db/conf/icassp/icassp2019.html#Kitrungrotsakul19

Publishing type：Research paper (scientific journal)  

DOI： 10.1109/TMM.2018.2877127

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1089/cmb.2018.0165

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/jcb/jcb26.html#YangHC19

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Association for Computing Machinery  

Liver cancer is one of the leading causes of death worldwide. Computer-aided diagnoses play an important role in liver lesion diagnoses (classification). Recently, several deep-learning-based computer-aided diagnosis systems have been proposed for the classification of liver lesions. The effectiveness of these systems has been demonstrated
however, the main challenge in deep-learning-based medical image classification is the lack of annotated training samples. In this paper, we demonstrate that transfer learning and fine-tuning can significantly improve the accuracy of liver lesion classification, especially for small training samples. We used the residual convolutional neural network (ResNet), which is a state-of-the-art network, as our baseline network for focal liver lesion classification using multiphase CT images. Fine-tuning significantly improved the classification accuracy from 83.7% to 91.2%. This classification accuracy (91.2%) is higher than that of state-of-the-art methods.

DOI： 10.1145/3299852.3299860

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：World Scientific Publishing Co. Pte Ltd  

Current image retrieval techniques are mainly based on text or visual contents. However, both text-based and contents-based methods lack the capability of utilizing human intuition and KANSEI (impression). In this paper, we proposed an impression-based image retrieval method in order to realize the image retrieval according to our impression presented by impression keywords. We first propose a generic and specific impressions estimation method based on machine learning and then apply it to impression-based clothing fabric image retrieval. We use a semantic differential (SD) method to measure the user's impressions such as brightness and warmth while they view a cloth fabric image. We also extract both global and local features of cloth fabric images such as color and texture using computer vision techniques. Then we use support vector regression to model the mapping functions between the generic impression (or specific impression) and image features. The learnt mapping functions are used to estimate the generic and specific impressions of cloth fabric images. The retrieval is done by comparing the query impression with the estimated impression of images in the database.

DOI： 10.1142/S0218001418540241

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

In this paper, we present a novel image representation approach for classifying emphysema in computed tomography (CT) images of the lung. Our proposed method extends rotation invariant uniform local binary pattern (RIULBP) and local ternary pattern (LTP), which are extensively used in a variety of computer vision applications, into rotation invariant uniform local ternary pattern (RIULTP) with a human perception principle: Weber's law. In addition, by integrating the upper pattern and the lower pattern of the Weber-based RIULTP (WRIULTP), we further put forward the joint Weber-based rotation invariant uniform local ternary pattern (JWRIULTP), which allows for a much richer representation and also takes the comprehensive information of the image into account. The proposed methods are tested on the Outex database (texture database) and the Bruijne and Sorensen database (emphysema database). The results show the superiority of the proposed approaches to the state-of-the-art techniques for emphysema classification including rotation invariant local binary pattern (RILBP) and texton-based approach.

DOI： 10.1109/ICIP.2017.8296642

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Part of collection (book)   Publisher：Springer Science and Business Media Deutschland GmbH  

This study addresses the classification problem of the HEp-2 cell using indirect immunofluorescent (IIF) image analysis, which can indicate the presence of autoimmune diseases by finding antibodies in the patient serum. Generally, the method used for IIF analysis remains subjective, and depends too heavily on the experience and expertise of the physician. Recently, studies have shown that it is possible to identify the cell patterns using IIF image analysis and machine learning techniques. However, it still has large gap in recognition rates to the physical experts’ one. This paper explores an approach in which the discriminative features of HEp-2 cell images in IIF are extracted and then, the patterns of the HEp-2 cell are identified using machine learning techniques. This study aims to realize a method for extracting highly-discriminant features from HEp-2 cell images by exploring a robust local descriptor inspired by Weber’s law. The investigated local descriptor is based on the fact that human perception for distinguishing a pattern depends not only on the absolute intensity of the stimulus but also on the relative variance of the stimulus. Therefore, we firstly transform the original stimulus (the images in our study) into a differential excitation-domain according to Weber’s law, and then explore a local patch, also called as micro-Texton, in the transformed domain as Weber local descriptor. Furthermore, we propose to employ a parametric probability process to model the Weber local descriptors, and extract the higher-order statistics to the model parameters for image representation. The proposed strategy can adaptively characterize the Weber local descriptor space using generative probability model, and then learn the parameters for better fitting the training space, which would lead to more discriminant representation for HEp-2 cell images. The simple linear support vector machine is used for cell pattern identification because of its low computational cost, in particular for large-scale datasets. Experiments using the open HEp-2 cell dataset used in the ICIP2013 contest validate that the proposed strategy can achieve a much better performance than the widely used local binary pattern (LBP) histogram and its extensions, Rotation Invariant Co-occurrence LBP (RICLBP) and Pairwise Rotation Invariant Co-occurrence LBP (PRICoLBP), and that the achieved recognition error rate is even very significantly below the observed intra-laboratory variability.

DOI： 10.1007/978-3-319-68843-5_6

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Science and Business Media Deutschland GmbH  

Segmentation of cellular and nuclear boundaries in differential interference contrast microscopy images is an important pre-processing step for biological image analysis. It is considered as a challenging task because of the interference of cell walls, blurs, nonuniform intensity background, and poor contrast between the foreground and the background. In this paper, we present a novel scheme on cellular boundary segmentation. Based on shape index (SI), the proposed method focuses on the detection of cytoplasm granules inside cellular regions. With several geometric post-processing techniques, the SI thresholding results are integrated into the segmented images. Because the size of the cytoplasm granules is usually too small comparing with the thickness of focal planes in Z-stack, we can not calculate SI values according to the method of constructing the intensity isosurface in 3D images. Consequently, we regard intensity as Z coordinate and compute SI values within each slice. A computed SI represents the shape of intensity surface or the variation of intensity near to a target pixel. Furthermore, we also show the proposed method can be applied to nuclear segmentation with a different post-processing step. Experimental results show the proposed algorithm has higher accuracy than existing schemes despite the existence of cell walls with different shapes and fluctuated intensities.

DOI： 10.1007/978-3-319-59397-5_8

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Science and Business Media Deutschland GmbH  

Graph cut is one of segmentation method that can give us the good result on natural image and large organ segmentation of medical image. However, we cannot get the correct or accurate results by using graph cut on detailed structures such as, tree branch, or blood vessel because the property of smoothness in graph cuts energy function will completely remove the small branch of the detailed structure to minimize its cost. We propose the vessel extraction method which combine graph cut and concept of submodular function. The conventional graph cuts will be use to obtain initial segmentation while graph cut with submodular function will be use to refine the initial segmentation. Submodular function can solve the problem of smoothness of graph cut in detail structure as shown in result that less segment and more united vessel tree than conventional graph cuts. The experimental result shows that our method can segment blood vessels of liver with higher accuracy while graph cut lead to a lot of loss of the detail branches in the liver vessel. With submodular constraint, we can connect the segment branch of vessel into united vessel tree which conventional graph cut still remain the segment of vessel’s branches.

DOI： 10.1007/978-3-319-59397-5_7

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Science and Business Media Deutschland GmbH  

There are multiple types of tumors occurring in the liver. Different tumors have different visual appearance and their visual appearance changes after injection of the contrast medium. So detection of liver tumors is considered as a challenging task. In this paper, we propose a method for detection of liver tumor candidates from CT images using a deep convolutional neural network. Experimental results show that we can significantly improve the detection accuracy by using our proposed method compared with the previous researches.

DOI： 10.1007/978-3-319-59397-5_15

Scopus

researchmap

Authorship：Lead author  

DOI： 10.1109/TIP.2018.2855418

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：Springer Verlag  

Purpose: The bag of visual words (BoVW) model is a powerful tool for feature representation that can integrate various handcrafted features like intensity, texture, and spatial information. In this paper, we propose a novel BoVW-based method that incorporates texture and spatial information for the content-based image retrieval to assist radiologists in clinical diagnosis. Methods: This paper presents a texture-specific BoVW method to represent focal liver lesions (FLLs). Pixels in the region of interest (ROI) are classified into nine texture categories using the rotation-invariant uniform local binary pattern method. The BoVW-based features are calculated for each texture category. In addition, a spatial cone matching (SCM)-based representation strategy is proposed to describe the spatial information of the visual words in the ROI. In a pilot study, eight radiologists with different clinical experience performed diagnoses for 20 cases with and without the top six retrieved results. A total of 132 multiphase computed tomography volumes including five pathological types were collected. Results: The texture-specific BoVW was compared to other BoVW-based methods using the constructed dataset of FLLs. The results show that our proposed model outperforms the other three BoVW methods in discriminating different lesions. The SCM method, which adds spatial information to the orderless BoVW model, impacted the retrieval performance. In the pilot trial, the average diagnosis accuracy of the radiologists was improved from 66 to 80% using the retrieval system. Conclusion: The preliminary results indicate that the texture-specific features and the SCM-based BoVW features can effectively characterize various liver lesions. The retrieval system has the potential to improve the diagnostic accuracy and the confidence of the radiologists.

DOI： 10.1007/s11548-017-1671-9

Scopus

researchmap

Publisher：SPIE  

DOI： 10.1117/12.2293074

researchmap

Authorship：Corresponding author   Publisher：IEEE Computer Society  

DOI： 10.1109/ISM.2018.000-7

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-319-92231-7_22

researchmap

Authorship：Lead author,　Corresponding author   Publisher：IEEE Computer Society  

DOI： 10.1109/ICPR.2018.8545634

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/FSKD.2018.8687125

researchmap

Publisher：IEEE  

DOI： 10.1109/GCCE.2018.8574753

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-319-97304-3_47

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-030-00934-2_74

researchmap

Publisher：IEEE  

DOI： 10.1109/ICIP.2018.8451514

researchmap

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-030-00767-6_64

researchmap

Authorship：Lead author,　Corresponding author   Publisher：Springer  

DOI： 10.1007/978-3-030-00767-6_17

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE  

Multimodal magnetic resonance images (e.g., T1-weighted image (TIWI) and T2-weighted image (T2WI)) are used for accurate medical imaging analysis. Different modal images have different resolution depending on pulse sequence parameters under limited data acquisition time. Therefore, interpolation methods are used to match the low-resolution (LR) image with the high-resolution (HR) image. However, the interpolation causes blurring that affects analysis accuracy. Although some recent works such as non-local-means (NLM) filter have manifested impressive super-resolution (SR) performance with available HR modal images, the filter has high computational cost. Therefore, we propose a fast SR framework with iterative-guided back projection, which incorporates iterative back projection with a guided filter (GF) method for resolution enhancement of LR images (e.g., T2WI) by referring HR images in another modality image (e.g., T1WI). The proposed method not only achieves both high accuracy than conventional interpolation methods and original GF and computational efficiency by applying an integral 3D image technique. In addition, although the proposed method is slightly inferior in accuracy visually than the state-of-the-art NLM filter, it can run 22 times faster than the state-of-theart method in expanding three times in the slice-select direction from 180 × 216 × 60 voxels to 180 × 216 × 180 voxels. The computational time of our method is about 1 min only. Therefore, the proposed method will be applied to various applications in practice, including not only multimodal MR images but also multimodal image analysis such as computed tomography (CT) and positron emission tomography (PET).

DOI： 10.1117/12.2285336

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Publisher：Springer  

DOI： 10.1007/978-3-030-00919-9_16

researchmap

Publisher：Springer  

DOI： 10.1007/978-3-030-00889-5_5

researchmap

Authorship：Lead author,　Corresponding author   Publisher：IEEE  

DOI： 10.1109/ICIP.2018.8451142

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Institute of Electrical and Electronics Engineers Inc.  

Content Based Image Retrieval (CBIR) systems that search similar images in a large database are attracting more and more research interests recently, and have been applied to medical image characterization for expert's experience sharing. One challenging task in CBIR is how to extract features for effective image representation. Therein sparse coding technique has been proven to be an effective way to learn inherent structure features for image analysis. However, it is necessary to first vectorize the 2- or 3-dimensional spatial structure for analysis with sparse coding, and then destroy the spatial relation of nearby voxels. In this study, we propose a multilinear sparse coding method to learn features from multi-dimensional medical images. We regard high dimensional local structures as tensors and propose a K-CP (CANDECOMP/PARAFAC) algorithm to learn a tensor dictionary in an iterative way. With the learned tensor dictionary, sparse coefficients of tensor local structures are calculated by multilinear orthogonal matching pursuit (MOMP) algorithm, which is an extended multilinear version of the conventional linear OMP. The proposed multilinear sparse coding method is prospected to be more efficient and effective for inherent feature extraction compared with conventional linear methods. The proposed method is applied to a CBIR system for retrieval of focal liver lesions (FLLs) using a medical database consisting of contrast-enhanced multi-phase computer-tomography (CT) images. Experiments show that the constructed CBIR with multilinear sparse coding method can achieve promising retrieval performance.

DOI： 10.1109/ISM.2017.100

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Institute of Electrical and Electronics Engineers Inc.  

Hyperspectral imaging has great potential for understanding the characteristics of different materials in many applications ranging from remote sensing to medical imaging. However, due to various hardware limitations, only low-resolution hyperspectral and high-resolution multi-spectral images can be available using existing imaging techniques. This study aims to generate a high-resolution hyperspectral image via fusion of the available LR-HS and HR-MS images. We propose a novel hyperspectral image superresolution method via non-negative sparse representation of reflectance spectral with adaptive sparsity constraint. By analyzing local content similarity of a focused pixel in the available high-resolution multi-spectral image, which can measure pixel material purity according to surrounding pixels, we generate a sparsity map for guiding non-negative sparse coding optimization procedure of the spectral representation called non-negative spectral representation with data-guided sparsity. Since the proposed method adaptively adjust the sparsity in the spectral representation based on the local content of the available high-resolution multi-spectral image, it can produce more robust spectral representation for recovering the target high-resolution hyper-spectral image. Comprehensive experiments on two public hyperspectral datasets validate that the proposed method achieves promising performances compared with the existing state of the art methods.

DOI： 10.1109/ISM.2017.99

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：MDPI AG  

Satellite scene classification is challenging because of the high variability inherent in satellite data. Although rapid progress in remote sensing techniques has been witnessed in recent years, the resolution of the available satellite images remains limited compared with the general images acquired using a common camera. On the other hand, a satellite image usually has a greater number of spectral bands than a general image, thereby permitting the multi-spectral analysis of different land materials and promoting low-resolution satellite scene recognition. This study advocates multi-spectral analysis and explores the middle-level statistics of spectral information for satellite scene representation instead of using spatial analysis. This approach is widely utilized in general image and natural scene classification and achieved promising recognition performance for different applications. The proposed multi-spectral analysis firstly learns the multi-spectral prototypes (codebook) for representing any pixel-wise spectral data, and then, based on the learned codebook, a sparse coded spectral vector can be obtained with machine learning techniques. Furthermore, in order to combine the set of coded spectral vectors in a satellite scene image, we propose a hybrid aggregation (pooling) approach, instead of conventional averaging and max pooling, which includes the benefits of the two existing methods, but avoids extremely noisy coded values. Experiments on three satellite datasets validated that the performance of our proposed approach is very impressive compared with the state-of-the-art methods for satellite scene classification.

DOI： 10.3390/ijgi6060175

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：ELSEVIER SCI LTD  

This study addresses the recognition problem of the HEp-2 cell using indirect immunofluorescent (IIF) image analysis, which can indicate the presence of autoimmune diseases by finding antibodies in the patient serum. Generally, the method used for IIF analysis remains subjective, and depends too heavily on the experience and expertise of the physician. This study aims to explore an automatic HEp-2 cell recognition system, in which how to extract highly discriminate visual features plays a key role in this recognition application. In order to realize this purpose, our main efforts include: (1) a simple but robust local descriptor without any quantization for local patch representation; (2)A transformation of the difference values between the surrounding pixels and the center one to the perception degree, which is based on the fact that human perception for disguising a pattern depends not only on the absolute intensity of the stimulus but also on the relative variance of the stimulus; called as Weber local descriptor (WLD); (3) a data-driven coding strategy with a parametric probability process, and the extraction of not only low- but also high-order statistics for image representation called as Fisher vector; (4) the stacking of the Fisher network into multi-layer framework for more discriminate feature. Experiments using the open HEp-2 cell dataset released in the ICIP2013 contest validate that the proposed strategy can achieve a much better performance than the state-of-the-art approaches, and that the achieved recognition error rate is even very significantly below the observed intra-laboratory variability.

DOI： 10.1016/j.patcog.2016.09.025

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

DOI： 10.1089/cmb.2016.0210

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/jcb/jcb24.html#YangHTKONC17

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE  

Extraction of blood vessels of the organ is a challenging task in the area of medical image processing. It is really difficult to get accurate vessel segmentation results even with manually labeling by human being. The difficulty of vessels segmentation is the complicated structure of blood vessels and its large variations that make them hard to recognize. In this paper, we present deep artificial neural network architecture to automatically segment the hepatic vessels from computed tomography (CT) image. We proposed novel deep neural network (DNN) architecture for vessel segmentation from a medical CT volume, which consists of three deep convolution neural networks to extract features from difference planes of CT data. The three networks have share features at the first convolution layer but will separately learn their own features in the second layer. All three networks will join again at the top layer. To validate effectiveness and efficiency of our proposed method, we conduct experiments on 12 CT volumes which training data are randomly generate from 5 CT volumes and 7 using for test. Our network can yield an average dice coefficient 0.830, while 3D deep convolution neural network can yield around 0.7 and multi-scale can yield only 0.6.

DOI： 10.1117/12.2253811

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：HINDAWI LTD  

Random walk (RW) method has been widely used to segment the organ in the volumetric medical image. However, it leads to a very large-scale graph due to a number of nodes equal to a voxel number and inaccurate segmentation because of the unavailability of appropriate initial seed point setting. In addition, the classical RW algorithm was designed for a user to mark a few pixels with an arbitrary number of labels, regardless of the intensity and shape information of the organ. Hence, we propose a prior knowledge-based Bayes random walk framework to segment the volumetric medical image in a slice-by-slice manner. Our strategy is to employ the previous segmented slice to obtain the shape and intensity knowledge of the target organ for the adjacent slice. According to the prior knowledge, the object/background seed points can be dynamically updated for the adjacent slice by combining the narrow band threshold (NBT) method and the organ model with a Gaussian process. Finally, a high-quality image segmentation result can be automatically achieved using Bayes RW algorithm. Comparing our method with conventional RW and state-of-the-art interactive segmentation methods, our results show an improvement in the accuracy for liver segmentation (p < 0 001).

DOI： 10.1155/2017/6506049

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE-INT SOC OPTICAL ENGINEERING  

Extraction of blood vessels of the organ is a challenging task in the area of medical image processing. It is really difficult to get accurate vessel segmentation results even with manually labeling by human being. The difficulty of vessels segmentation is the complicated structure of blood vessels and its large variations that make them hard to recognize. In this paper, we present deep artificial neural network architecture to automatically segment the hepatic vessels from computed tomography (CT) image. We proposed novel deep neural network (DNN) architecture for vessel segmentation from a medical CT volume, which consists of three deep convolution neural networks to extract features from difference planes of CT data. The three networks have share features at the first convolution layer but will separately learn their own features in the second layer. All three networks will join again at the top layer. To validate effectiveness and efficiency of our proposed method, we conduct experiments on 12 CT volumes which training data are randomly generate from 5 CT volumes and 7 using for test. Our network can yield an average dice coefficient 0.830, while 3D deep convolution neural network can yield around 0.7 and multi-scale can yield only 0.6.

DOI： 10.1117/12.2253811

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：Hindawi Publishing Corporation  

Characterization and individual trait analysis of the focal liver lesions (FLL) is a challenging task in medical image processing and clinical site. The character analysis of a unconfirmed FLL case would be expected to benefit greatly from the accumulated FLL cases with experts' analysis, which can be achieved by content-based medical image retrieval (CBMIR). CBMIR mainly includes discriminated feature extraction and similarity calculation procedures. Bag-of-Visual-Words (BoVW) (codebook-based model) has been proven to be effective for different classification and retrieval tasks. This study investigates an improved codebook model for the fined-grained medical image representation with the following three advantages: (1) instead of SIFT, we exploit the local patch (structure) as the local descriptor, which can retain all detailed information and is more suitable for the fine-grained medical image applications
(2) in order to more accurately approximate any local descriptor in coding procedure, the sparse coding method, instead of K-means algorithm, is employed for codebook learning and coded vector calculation
(3) we evaluate retrieval performance of focal liver lesions (FLL) using multiphase computed tomography (CT) scans, in which the proposed codebook model is separately learned for each phase. The effectiveness of the proposed method is confirmed by our experiments on FLL retrieval.

DOI： 10.1155/2017/1413297

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

Focal liver lesion enhancement and detection has an essential role for the computer-aided diagnosis and characterization of lesion regions in CT volume data. This paper proposes a novel focal lesion enhancement strategy by extracting a lesion saliency map, which represents the deviation degree of the uncommon or lesion tissue from the common tissues (liver and vessel) in CT volumes. The saliency map can be constructed by exploring the existing probability of lesion for any voxel. However, due to the large diversity of liver lesions, it is difficult to construct an universal model for all types of lesions. Therefore, this study proposes to construct probability models of the common tissues, which have comparably small variability even for different samples and is relatively easy to obtain the prototype regions even from the understudying CT volume. In order to robustly and flexibly characterize the common tissues, we explore a Bayesian framework by combining a general model, which is constructed oriented to all CT samples, and an adaptive model, which is constructed specific to the under-studying CT sample, for calculating the existing probability of the common tissues (liver or vessel). Then, the saliency map (the existing probability) of focal lesion can be deduced from that of liver or vessel. The advantages of our proposed strategy mainly include three aspects: (1) it only needs to prepare the prototypes of common tissue such as liver or vessel region, which are easily obtained in any CT liver volume
(2) it proposes to combine the general and adaptive model as Bayesian framework for more robust and flexible characterization of the common tissue
(3) dispensable to remove the other different structure such as vessel in liver volume as a pre-processing step. Experiments validate that the proposed Bayesian-based saliency model for focal liver lesion enhancement can perform much better than the conventional approaches such as EM, EM/MPM based lesion detection and segmentation methods.

DOI： 10.1007/978-3-319-54526-4_3

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：ELSEVIER SCIENCE BV  

In this paper, we describe a novel image representation strategy for classifying HEp-2 cell patterns of fluorescence staining. Our proposed strategy extends local binary patterns (LBPs), which are state-of-the-art texture features, into local ternary patterns (LTPs) with data-driven thresholds according to Weber's law, a human perception principle; further, our approach incorporates the contexts of spatial and orientation co-occurrences among adjacent Weber-based local ternary patterns (WLTPs) for texture representation. The explored WLTP is formulated by adaptively quantizing differential values between neighborhood pixels and the focused pixel as negative or positive stimuli if the normalized differential values are large; otherwise the stimulus is set to 0. Our approach here is based on the fact that human perception of a distinguished pattern depends not only on the absolute intensity of the stimulus but also on the relative variance of the stimulus. By integrating spatial and orientation context information, we further propose a rotation invariant co-occurrence WLTP (RICWLTP) approach to be more discriminant for image representation. Through experiment on the open HEp-2 cell dataset used at the ICPR2014 contest, we confirmed that our proposed strategy can greatly improve recognition performance or achieve comparable performance as compared with state-of-the-art LBP-based descriptor, the conventional LTP, and adaptively codebook/model based methods. (C) 2016 Elsevier B.V. All rights reserved.

DOI： 10.1016/j.patrec.2016.02.004

J-GLOBAL

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

The polarization angle-independent, dual-band, 90 degrees polarization rotator presented in this paper consisted of a bilayered chiral metamaterial composed of twisted capacity-loaded I-shaped electric field-coupled resonators in C4 symmetry. Simulated and measured results consistently demonstrated that the rotator exhibited extremely low loss and high polarization ratio under dual-band conditions. We also systematically investigated the dependence of the electromagnetic response of the loaded structure on the geometric parameters. The proposed simple, easily fabricated model of a chiral metamaterial may be used in further polarization rotator applications.

DOI： 10.1587/elex.13.20160583

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1186/s41074-016-0010-3

researchmap

Other Link： https://dblp.uni-trier.de/db/journals/ipsjtcva/ipsjtcva8.html#HanTKONC16

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE  

Probabilistic atlas based on human anatomical structure has been widely used for organ segmentation. The challenge is how to register the probabilistic atlas to the patient volume. Additionally, there is the disadvantage that the conventional probabilistic atlas may cause a bias toward the specific patient study due to a single reference. Hence, we propose a template matching framework based on an iterative probabilistic atlas for organ segmentation. Firstly, we find a bounding box for the organ based on human anatomical localization. Then, the probabilistic atlas is used as a template to find the organ in this bounding box by using template matching technology. Comparing our method with conventional and recently developed atlas-based methods, our results show an improvement in the segmentation accuracy for multiple organs (p &lt
0:00001).

DOI： 10.1117/12.2217340

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we describe a novel image representation strategy for classifying HEp-2 cell patterns of fluorescence staining. Our proposed strategy extends local binary patterns (LBPs), which are state-of-the-art texture features, into local ternary patterns (LTPs) with data-driven thresholds according to Weber's law, a human perception principle; further, our approach incorporates the contexts of spatial and orientation co-occurrences among adjacent Weber-based local ternary patterns (WLTPs) for texture representation. The explored WLTP is formulated by adaptively quantizing differential values between neighborhood pixels and the focused pixel as negative or positive stimuli if the normalized differential values are large; otherwise the stimulus is set to 0. Our approach here is based on the fact that human perception of a distinguished pattern depends not only on the absolute intensity of the stimulus but also on the relative variance of the stimulus. By integrating spatial and orientation context information, we further propose a rotation invariant co-occurrence WLTP (RICWLTP) approach to be more discriminant for image representation. Through experiment on the open HEp-2 cell dataset used at the ICIP2013 contest, we confirmed that our proposed strategy can greatly improve recognition performance or achieve comparable performance as compared with state-of-the-art LBP-based descriptor, the conventional LTP, and adaptively codebook/model based methods.

DOI： 10.1007/978-3-319-23024-5_19

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

This paper proposes a novel framework for fully automatic localization of deformable organs in medical volume data, which can obtain not only the position but also simultaneously the orientation and deformation of the organ to be searched, without the need to segment the organ first. The problem is defined as one of minimizing the sum of squared distances between the organ model's surface points and their closest surface points extracted from the input volume data. The geometric alignment, or so-called registration, of three-dimensional models by least square minimization always has the problem of initial states. We argue that the only way to solve this problem is by the exhaustive search. However, the exhaustive search takes much computational cost. In order to reduce the computational cost, we make efforts in the following three ways: (1) a uniform sampling over 3D rotation group; (2) Pyramidal search for all parameters; (3) Construction of a distance function for efficiently finding closest points. We have finished experiments for searching the six parameters for position and orientation, and the results show that the proposed framework can achieve correct localization of organs in the input data even with very large amounts of noise. We are currently expanding the system to localize organs with large deformation by adding and searching parameters representing scaling and deformation.

DOI： 10.1007/978-3-319-23024-5_28

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER INT PUBLISHING AG  

Automatic liver lesion enhancement and detection has an essential role for the computer-aided diagnosis of liver tumor in CT volume data. This paper proposes a novel lesion enhancement strategy using Bayesian framework by combining the lesion probabilities based on an adaptive non-parametric model with the processed test volume and the constructed common non-lesion models with prepared liver database. Due to the large variation of different lesion tissues, it is difficult to obtain the common lesion prototypes from liver volumes, and thus this paper investigates a lesion-training-data free strategy by only constructing the healthy liver and vessel prototypes using local patches, which can be extracted from any slice of the test liver volume, and is also easy to prepare the common training non-lesion samples for all volumes. With the healthy liver and vessel prototypes from the test volume, an adaptive non-parametric model is constructed for estimating the lesion possibility, which is considered as the pixel likelihood to lesion region; the common model constructed using the pre-prepared liver database is used to estimate the pixel probability, which is defined as prior knowledge due to the used unvaried model. Finally, the posterior probabilities based on Bayesian theory are achieved for enhancing lesion regions. Experimental results validate that the proposed framework can not only detect almost small lesion regions but also greatly reduce falsely detect regions.

DOI： 10.1007/978-3-319-39687-3_22

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

Recently, content-based image retrieval (CBIR) in medical applications has attracted a lot of attentions. In this paper, we present a preliminary study on CBIR of focal liver lesions based on combined density, texture and shape features of multi-phase contrast-enhanced CT volumes. We improve the existing method from following two aspects: (1) in order to improve the retriev al accuracy, we propose a novel 3D shape feature for CBIR of liver lesions in addition to conventional density and texture features; (2) in order to reduce the computation time, we propose an improved local binary pattern, which is called imLBP, as the 3D texture feature. The effectiveness of our proposed method has been validated with real clinical datasets.

DOI： 10.1007/978-3-319-23024-5_20

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

A fast component-counting algorithm is proposed based on graph theory in this paper. We derived a formulation to count faces in a plane given only the vertices based on Euler polyhedron formula. Vertices with degree no more than two are ineffective in counting components. With the derived formula, a graph component counting algorithm is constructed based only on searching cross points whose degree are no less than three. When applied to a two-dimensional binary image, the proposed method divides an image into patches of same size and decides which of them will be used in counting by searching the circumferential pixels of each patch. If the number of component edges within the circumferential pixels of a patch is no less than three, then the patch will be used in counting. After determining all the vertices with degree no less than three, the number of components can be calculated by the formula. Because only a small number of pixels are investigated in the process, the computational time is very fast. The disconnection of edges in an image is one of the main reasons which causes miscounts for scanning-based algorithms. The difficulty, however, can be naturally overcome by the proposed algorithm because disconnected points will be identified as futile pixels in the algorithm. Experimental results show the algorithm is more efficient than existing methods. When applied to images with disconnected edges, the counted number given by scanning-based algorithms is much smaller than the correct number whereas the proposed algorithm obtains satisfactory results.

DOI： 10.1109/SNPD.2016.7515972

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE-INT SOC OPTICAL ENGINEERING  

Probabilistic atlas based on human anatomical structure has been widely used for organ segmentation. The challenge is how to register the probabilistic atlas to the patient volume. Additionally, there is the disadvantage that the conventional probabilistic atlas may cause a bias toward the specific patient study due to a single reference. Hence, we propose a template matching framework based on an iterative probabilistic atlas for organ segmentation. Firstly, we find a bounding box for the organ based on human anatomical localization. Then, the probabilistic atlas is used as a template to find the organ in this bounding box by using template matching technology. Comparing our method with conventional and recently developed atlas-based methods, our results show an improvement in the segmentation accuracy for multiple organs (p < 0.00001).

DOI： 10.1117/12.2217340

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER INT PUBLISHING AG  

This study addresses the recognition problem of the HEp-2 cell using indirect immunofluorescent (IIF) image analysis, which can facilitate the diagnosis of many autoimmune diseases by finding antibodies in the patient serum. Recently, a lot of automatic HEp-2 cell classification strategies including both shallow and deep methods have been developed, wherein the deep Convolutional Neural Networks (CNNs) have been proven to achieve impressive performance. However, the deep CNNs in general requires a fixed size of image as the input. In order to conquer the limitation of the fixed size problem, a spatial pyramid pooling (SPP) strategy has been proposed in general object recognition and detection. The SPP-net usually exploit max pooling strategies for aggregating all activated status of a specific neuron in a predefined spatial region by only taking the maximum activation, which achieved superior performance compared with mean pooling strategy in the traditional state-of-the-art coding methods such as sparse coding, linear locality-constrained coding and so on. However, the max pooling strategy in SPP-net only retains the strongest activated pattern, and would completely ignore the frequency: an important signature for identifying different types of images, of the activated patterns. Therefore, this study explores a generalized spatial pooling strategy, called K-support spatial pooling, in deep CNNs by integrating not only the maximum activated magnitude but also the response magnitude of the relatively activated patterns of a specific neuron together. This proposed K-support spatial pooling strategy in deep CNNs combines the popularly applied mean and max pooling methods, and then avoid awfully emphasizing of the maximum activation but preferring a group of activations in a supported region. The deep CNNs with the proposed K-support spatial pooling is applied for HEp-2 cell classification, and achieve promising performance compared with the state-of-the-art approaches.

DOI： 10.1007/978-3-319-46976-8_1

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

Computer-aided diagnosis (CAD) systems have been verified to have the potential to assist radiologists in clinical diagnosis to detect and characterize focal liver lesions (FLLs) based on single-or multiphase contrast-enhanced computed tomography (CT) images. Features extracted from multiphase contrast-enhanced CT images carry more important diagnostic information i.e. enhancement pattern and demonstrate much stronger discriminative ability compared to those of single-phase CT images. In this paper, we propose a new method for multiphase image feature generation called the bag of temporal co-occurrence words (BoTCoW). A temporal co-occurrence image connecting intensity from multiphase images is constructed. Then the bag of visual word (BoVW) model is employed on the temporal co-occurrence images to extract temporal features. The proposed method effectively captures temporal enhancement information and demonstrates the distribution of the evolution patterns. The effectiveness of this method is validated in a retrieval system using 132 FLLs with confirmed pathology type. The preliminary results show that the proposed BoTCoW method outperforms the previously proposed temporal features and multiphase features based on the BoVW model.

DOI： 10.1109/ICPR.2016.7899976

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper presents super resolution (SR) of 3D MR images which is effective for brain segmentation as practical application. Brain segmentation is one of important tasks to analyze brain morphometry. An accurate brain segmentation helps improve accuracy of post-processing. The segmentation is affected by the resolution of 3D MR images. In particular, the resolution in slice-select direction is much lower than the in-plane direction in some type of 3D MR images (e.g. T2-weighted images and proton density images). Therefore, we apply a learning-based SR method using sparse representation and self-similarity for generating high-resolution (HR) in slice-select direction to be same resolution in in-plane direction. With the visualized evaluation, it can be seen that the segmentation accuracy in SR results are improved by reducing partial volume effect. With the quantitative evaluation, one can confirm that the DICE value of SR results are higher than conventional interpolation method.

DOI： 10.1109/CISP-BMEI.2016.7852827

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In this paper, we propose a SIFT-based multi-frame super resolution for 250 million pixel images. In the proposed method, we first use the SIFT operator to detect key points in each frame. Then we use a closest matching method to find the correspondence among multi-frame images. The corresponding key points are used to register multi-frame images to a reference image, which is randomly selected from the multi-frame images. After registration, we combine the aligned multi-frame images to form a high-quality and high-resolution image. We applied the proposed method to enhance the quality of 250 million pixel images, which is obtained by the Canon's 250Mpixel CMOS-image-sensor.

DOI： 10.1109/CISP-BMEI.2016.7852826

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Food is one of the most important issues for human health. In order to manage the healthy dietary life, we are going to develop an auto food-log record system, which is based on auto food recognition technologies. To realize this system, we aim to propose a discriminated food image representation that can perform effective identification of food images in this paper. The conventional image representation mainly includes color and texture distributions (histogram), which are the statistical information based on uniformly quantized color or texture levels. However, these conventional techniques using uniform quantization of the on-hand color and texture in the image lead much information loss for reliably constructing the image. Therefor, this study proposes to characterize the color and texture information by incorporating the strategy of patch-based bag of features model. This technique can adaptively learn the representative color or texture (prototypes) from the food images for food recognition, and it is possible to recover a more reliable image using the learned prototypes. The experiments using our proposed approaches show that the recognition rate can be greatly improved compared with the conventional method.

DOI： 10.1109/CISP-BMEI.2016.7852822

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Content-based image retrieval (CBIR) technique for digital image searching has been applied in medical images which are called content-based medical image retrieval (CBMIR). In this paper, we combine texture, density and shape features for CBMIR based on 3D multi-phase contrast enhanced CT images according to radiologists' clinical experience. And a database of 132 focal liver lesions (FLLs) with confirmed pathology type is constructed in this work. We implement bag of visual words (BoVW) model to extract texture features from FLLs based on 3D local binary pattern (LBP) and combine it with conventional intensity-based BoVW. Density and temporal density are designed based on clinical observation. Principle Component Analysis (PCA) is used to extract the sphericity of the lesions as shape features. Experiments performed on the FLLs database show the efficiency of the proposed system and features.

DOI： 10.1109/CISP-BMEI.2016.7852820

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.2197/ipsjjip.24.320

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IOS PRESS  

The automatic inspection of throw-away tips is very important for quality control in precision cutting. We proposed an image processing based method for automatic inspection of the processing wear of throw-away tips. After image denoising, the proposed method utilized image-patch based principal component analysis method to enhance the cutting worn region while suppress the background region. Then the enhanced worn region was automatically segmented by a simple thresholding method followed by post-processing. The area of the segmented worn region was used as a measure of cutting wear degree. We collected three datasets of time-series images that recorded the processing of throw-away tips on a product line. One dataset was used to choose optimal parameters of the proposed method, and the other two datasets were used for evaluate its performances. Experimental results showed that the proposed method was able to inspect the cutting wear with high accuracy. Additionally, it was also showed that the proposed method outperformed the conventional thresholding based method.

DOI： 10.3233/JIFS-169020

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：PERGAMON-ELSEVIER SCIENCE LTD  

Purpose: Assessing the treated region with locoregional therapy (LT) provides valuable information for predicting hepatocellular carcinoma (HCC) recurrence. The commonly used of assessment method is inefficient because it only compares two-dimensional CF images manually. In our previous work, we automatically aligned the two CT volumes to evaluate the therapeutic efficiency using registration algorithms. The non-rigid registration is applied to capture local deformation, however, it usually destroys internal structure. Taking these into consideration, this paper proposes a novel non-rigid registration approach for evaluating LT of HCC to maintain the image integrity.
Method: In our registration algorithm, a global affine transformation combined with localized cubic B-spline is used to estimate the significant non-rigid motions of two livers. The proposed method extends a classical non-rigid registration based on mutual information (MI) that uses an anatomical structure term to constrain the local deformation. The energy function can be defined based on the total one associated with the anatomical structure and deformation information. Optimal transformation is obtained by finding the equilibrium state in which the total energy is minimized, indicating that the anatomical landmarks have found their correspondences. Thus, we can use the same transformation to automatically transform the ablative region to the optimal position.
Results: Registration accuracy is evaluated using the clinical data. Improved results are obtained with respect to all criteria in our proposed method (MI-LC) than those in the MI-based non-rigid registration. The landmark distance error (LDE) of MI-LC is decreased by an average of 3.93 mm compared to the case of MI-based registration. Moreover, it could be found regardless of how many landmarks applied in our proposed method, a significant reduction in LDE values using registrations based on MI-LC compared with those based on MI is confirmed.
Conclusion: Our proposed approach can guarantee the continuity, the accuracy and the smoothness of structures by constraining the anatomical features. The results clearly indicate that our method can retain the local deformation of the image. In addition, it assures the anatomical structure stability. (C) 2015 Elsevier Ltd. All rights reserved.

DOI： 10.1016/j.compmedimag.2015.08.003

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Organ segmentation is one of the most fundamental and challenging tasks in computer aid diagnosis system. Researches successfully working on interactive segmentation for medical image include graph cuts and random walk. However, graph cut-based organ segmentation for 3D medical volume data requires an optimization procedure of cutting the object/background regions on a very largescale graph, which not only consumes large amount of memory and but also requires expensive computational cost. This paper conquers the drawbacks of graph cut-based organ segmentation via instead of voxel with supervoxel as nodes for constructing graph in interactive 3D organ segmentation that can greatly reduces node number and connected edges
voxels of medical data with similar intensity magnitude and near spatial relation are grouped into the supervoxels and such supervoxels are used as pseudo nodes of graph for cutting object (organ) and background regions, named as supervoxel-based graph cut. To validate the effectiveness and efficiency of the proposed method, we conduct experiments on 10 medical data, which possibly have tumors inside organ, or have abnormal deformed organ shape. The experimental results show that the proposed method is much superior than conventional graph cutbased method in term of accuracy, computational time and memory usage.

DOI： 10.1016/j.ifacol.2015.10.117

Scopus

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC  

Highly discriminant visual features play a key role in different image classification applications. This study aims to realize a method for extracting highly-discriminant features from images by exploring a robust local descriptor inspired by Weber's law. The investigated local descriptor is based on the fact that human perception for distinguishing a pattern depends not only on the absolute intensity of the stimulus but also on the relative variance of the stimulus. Therefore, we firstly transform the original stimulus (the images in our study) into a differential excitation-domain according to Weber's law, and then explore a local patch, called micro-Texton, in the transformed domain as Weber local descriptor (WLD). Furthermore, we propose to employ a parametric probability process to model the Weber local descriptors, and extract the higher-order statistics to the model parameters for image representation. The proposed strategy can adaptively characterize the WLD space using generative probability model, and then learn the parameters for better fitting the training space, which would lead to more discriminant representation for images. In order to validate the efficiency of the proposed strategy, we apply three different image classification applications including texture, food images and HEp-2 cell pattern recognition, which validates that our proposed strategy has advantages over the state-of-the-art approaches.

DOI： 10.1109/TCYB.2014.2346793

J-GLOBAL

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：WILEY-BLACKWELL  

In recent years, there has been increasing interest in statistical shape modeling of human anatomy. The statistical shape model can capture the morphological variations of human anatomy. Since liver cirrhosis will cause significant morphological changes, the authors propose a computer-aided diagnosis method for liver cirrhosis based on statistical shape models. In the proposed method, the authors first construct a statistical shape model of the liver using 50 clinical CT datasets (25 sets of normal data and 25 sets of abnormal data). The authors apply the marching cubes algorithm to convert the segmented liver volume to a triangulated mesh surface containing 1000 vertex points. The coordinates of these vertex points are used to represent the 3D liver shape as a shape vector. After normalization and identification of correspondences between all datasets, principal component analysis (PCA) is employed to find the principal variation modes of the shape vectors. Then the authors propose a mode selection method based on class variations between the normal class and abnormal class. The authors found that the top two modes of class variations are most effective for the classification of normal and abnormal livers. The classification rate of abnormal livers and normal livers by the use of a simple linear discriminant function were 84% and 80%, respectively. (C) 2014 by Wiley Periodicals.

DOI： 10.1002/eej.22668

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Glycosylation Site Prediction (GSP) research has witnessed a growing interest in proteomics. The high ability to GSP is helpful for better understanding the function of protein, theoretically. In this research, our aim is to explore a new method for improving the performance of GSP of O-glycosylation sites. We propose to utilize Independent Component Analysis (ICA) for feature selection and dimension reduction, and then use Support Vector Machine (SVM) for glycosylation site classification, in which our method is applied for two kinds of datasets in glycosylated site and nonglycosylated site. Compared with using other subspace-based method and SVM method, experimental results show that our new approach is feasible and effective with higher prediction accuracy.

researchmap

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：JAPAN SOC SIMULATION TECHNOLOGY-JSST  

Ubiquitous computing is an important technology in medicine that is predicted to support doctors anywhere and anytime. To help achieve it, this paper develops the Interactive Segmentation and Visualization System for Medical images on Mobile devices (ISVS_M-2), which originally was designed to work on workstations, but also works on a wide range of mobile devices via a mobile client-server platform. The developed ISVS_M-2 basically consists of three modules: asegmentation module that is implemented on a server; commutation modules on both the server and mobile device; and interactive and visualization modules on the mobile device that not only give visualization of internal information of an organ model, but also interactively refine organ segmentation according to user experience. With interaction via a computer graphic interface on the mobile device, and communication via the mobile client-server platform, ISVS_M-2 offers users a novel and efficient approach to computer-aided medicine.

DOI： 10.15748/jasse.2.96

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1016/j.compbiomed.2015.10.007

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In this paper, we describe a novel local descriptor of image texture representation for visual recognition. The image features based on micro-descriptors such as local binary patterns (LBP) and local ternary patterns (LTP) have been very successful in number of applications including face recognition and texture analysis. Instead of binary quantization in LBP, LTP thresholds the differential values between a focused pixel and its neighborhood pixels into three graylevel, which can be explained as the active status (i.e., positively activated, negatively activated and not activated) of the neighborhood pixels compared to the focused pixel. However, regardless to the magnitude of the focused pixel, the thresholding strategy remains fixed, which would violate the principle of human perception. Therefore, in this study, we design LTP with a data-driven threshold according to Weber's law, a human perception principle; further, our approach incorporates the contexts of spatial and orientation co-occurrences (i.e., co-occurrence context) among adjacent Weber-based local ternary patterns (WLTPs) for texture representation. In order to validate efficiency of our proposed strategy, we apply to three different visual recognition applications including two texture datasets and one food image dataset, and prove the promising performance can be achieved compared with the state-of-the-art approaches.

researchmap

Authorship：Lead author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

This study addresses the recognition problem of the HEp-2 cell using indirect immunofluorescent (IIF) image analysis, which can indicate the presence of autoimmune diseases by finding antibodies in the patient serum. Generally, the method used for IIF analysis remains subjective, and depends too heavily on the experience and expertise of the physician. This study aims to explore an automatic HEp-2 cell recognition system, in which how to extract highly discriminate visual features plays a key role in this recognition application. In order to realize this purpose, our main efforts include: (1) a transformed excitation domain instead of the raw image domain, which is based on the fact that human perception for disguising a pattern depends not only on the absolute intensity of the stimulus but also on the relative variance of the stimulus
(2) a simple but robust micro-texton without any quantization in the excitation domain, called as Weber local descriptor (WLD)
(3) a data-driven coding strategy with a parametric probability process, and the extraction of not only low- but also high-order statistics for image representation called as Fisher vector
(4) the stacking of the Fisher network into deep learning framework for more discriminate feature. Experiments using the open HEp-2 cell dataset released in the ICIP2013 contest validate that the proposed strategy can achieve a much better performance than the state-of-the-art approaches, and that the achieved recognition error rate is even very significantly below the observed intra-laboratory variability.

DOI： 10.1007/978-3-319-24888-2_11

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Liver segmentation is one of the most fundamental and challenging tasks in computer aided diagnosis (CAD) system for liver diseases. Graph cut algorithms have been successfully applied to medical image segmentation of different organs for 3D volume data, which not only leads to very large-scale graph due to the same node number as voxel number, but also completely ignore some available organ shape priors. Thus, a slice by slice liver segmentation method by combining shape constraints according to previously slice segmentation has been proposed based on graph cut. However, the constructed graph scale is still large, and the computation of distance map from all voxel to the segmented shape leads to high cost. In order to explore an efficient and effective slice by slice segmentation method for liver, this paper proposes to apply clustering algorithm to firstly group slice pixels into superpixels as nodes for constructing graph, which not only greatly reduce the graph scale but also significantly speed up the optimization procedure of the graph. Furthermore, we restrict the regions near organ boundary as shape constraints, which can further reduce computational time. To validate effectiveness and efficiency of our proposed method, we conduct experiments on 10 CT volumes, most of which have tumors inside liver, and abnormal deformed shape of liver. Our method can yield an average dice coefficient: 0.94, about 659.22 second in computation, and take only 1.5GB in memory usage.

DOI： 10.1109/ICIP.2015.7351428

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In medical diagnosis, high resolution (HR) images are indispensable for giving more correct decision. The super resolution technique, which can generate HR images from LR images based on machine learning, attracts hot attention recently. However, the conventional learning based SR generally cannot recover high frequency information. In this paper, we integrate a further learning step into the conventional method, and proposes a two-step learning based SR, which is prospected to recover most high frequency information lost in the available LR input. Furthermore, we also propose to use HR axial plane images of input volumes as HR training data to reconstruct HR coronal plane and sagittal plane images.

DOI： 10.1109/GCCE.2015.7398738

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Consumers' psychological feeling or impression is an important factor for product design. The impression estimation becomes an important issue. In this paper, we propose generic and specific impression estimation methods based on machine learning for cloth fabric images. We use a semantic differential (SD) method to measure the user's impression such as bright, warm while they viewing a cloth fabric image. We also extract both global and local features of cloth fabric images such as color and texture using computer vision techniques. Then we use support vector regression to model the mapping functions between the generic impression (or specific impression) and image features. The learned mapping functions are used to estimate the generic or specific impression of cloth fabric images.

DOI： 10.1109/FSKD.2015.7382212

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

The automatic inspection of throw-away tips is very important for quality control in precision cutting. We proposed an image processing based method for automatic inspection of the processing wear of throw-away tips. Our proposed method utilized principal component analysis as a preprocessing to reduce the noise and enhance the contrast of the cutting wear region with the background region. Then the PCA transformed wear region is automatically segmented by a simple thresholding method. The area of the segmented wear region is used as a measure of cutting wear degree.

DOI： 10.1109/FSKD.2015.7382210

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

A random walks-based (RW) segmentation method has been gaining popularity in recent years with its ability to interactively segment the objects with minimal guidance. It has potential applications in segmenting the 3D image. However, due to the large computational burden of the classical RW algorithm, it is a challenge to use this algorithm to segment 3D medical images interactively. Hence, a knowledge-based segmentation framework for the liver is proposed based on random walks and narrow band threshold (RWNBT). Our strategy is to employ the previous segmented slice to achieve a prior knowledge (the shape and intensity constraints) of liver for automatic segmentation of the adjacent slice. With a small number of user-defined seeds, we can obtain the segmentation results of the start slice in the volume which would be used as the prior knowledge of the segmented organ. According to this intensity constraints, the "Candidate Pixels" image can be generated by thresholding the organ models with Gaussian Mixture Model (GMM), which can remove the noise and non-liver parts. Furthermore, the object/background seeds can be dynamically updated for the adjacent slice by combining a narrow band threshold (NBT) method and the shape constrains. Finally, a combinational random walker algorithm is applied to automatically segment the whole volume in a slice-by-slice manner. Comparing our method with conventional RW and the state-of-the-art interactive segmentation methods, our results show an improvement in the accuracy for liver segmentation.

DOI： 10.1109/FSKD.2015.7382208

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper proposes a robust registration and alignment framework for multiple point clouds using low rank and sparse decomposition. A coarse registration phase utilizing Huber-ICP is firstly performed to roughly align all the point clouds to a same location, and then sparse and low rank decomposition is applied to extract the low rank subspace of all the point clouds, which is expected to be outlier and loss data free. Finally, a fine registration procedure can be carried out between each point clouds from this low rank space to not only a more accurate registration result but also a more precise correspondence. Robustness of our method for outliers contained in point clouds is verified through manufactured data and it also shows that an effective result can still be achieved even when some points in the cloud are lost.

DOI： 10.1109/APSIPA.2015.7415371

researchmap

Authorship：Lead author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In this paper, we describe a novel local descriptor of image texture representation for visual recognition. The image features based on micro-descriptors such as local binary patterns (LBP) and local ternary patterns (LTP) have been very successful in number of applications including face recognition and texture analysis. Instead of binary quantization in LBP, LTP thresholds the differential values between a focused pixel and its neighborhood pixels into three graylevel, which can be explained as the active status (i.e., positively activated, negatively activated and not activated) of the neighborhood pixels compared to the focused pixel. However, regardless to the magnitude of the focused pixel, the thresholding strategy remains fixed, which would violate the principle of human perception. Therefore, in this study, we design LTP with a data-driven threshold according to Weber's law, a human perception principle; further, our approach incorporates the contexts of spatial and orientation co-occurrences (i.e., co-occurrence context) among adjacent Weber-based local ternary patterns (WLTPs) for texture representation. In order to validate efficiency of our proposed strategy, we apply to three different visual recognition applications including two texture datasets and one food image dataset, and prove the promising performance can be achieved compared with the state-of-the-art approaches.

DOI： 10.1109/ACPR.2015.7486617

researchmap

Authorship：Lead author,　Corresponding author   Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Nuclear detection in embryos is an indispensable process for quantitative analysis of the development of multicellular organisms. Due to the overlap in the distribution of nuclear and cytoplasmic intensities and the large variation even within the same type of tissues of different embryos, it is difficult to separate nuclear regions from the surrounding cytoplasmic region in differential interference contrast (DIC) microscope image. This study explores a discriminative representation of texton around a fixed pixel, called Top-ranked Intensity-ordered Descriptor (TRIOD), which is prospected to distinguish the smoothed texture in nucleus from the irregular texture in cytoplasm containing yolk granules. Then, a probability process is employed to model nuclear TRIOD prototypes, and the enhanced nuclear probability map can be constructed with the TRIODs of all pixels in a DIC microscope image. Finally, distance regularized level set method is applied to refine the initial localization by simply thresholding on the enhanced probability map. Experimental results show that the proposed strategy can give much better performance for segmentation of nuclear regions than the conventional strategies.

DOI： 10.1109/ACPR.2015.7486564

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

The residences of Japanese Archipelago mainly include the three human populations; the Ainu, the Mainland Japanese and the Ryukyuan, which can he inferred by genome-wide single-nucleotide polymorphism (SNP) data and characterized by generic base sequences. In the other hand, the genetic association of human :facial morphological variation recently becomes a more and more active research field, aims to quantitatively analyze the possible relation measure between the gene base and a kind of facial morphological variations. This study attempts to explore the discriminated phenotype features of the common facial morphological variations between the Mainland Japanese and the Ryukyuan; the difference of phenotype features between these two populations is prospected to infer different gene base sequences. In order to explore the facial phenotype features, we propose a framework of local statistical analysis for adjacent geometrical regions of 3D facial images. Therein, we firstly select the surface points with higher distinguishable values based fisher linear discriminate analysis as discriminated coordinate vectors, and further cluster them into local geometrical groups for morphological analysis. The extracted local phenotype features are applied for identification of two populations, and achieve the comparable or better performances than the global phenotype feature, which manifests the possibility fOr association analysis between local morphological phenotype and the genes.

DOI： 10.1109/ACPR.2015.7486524

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

Chronic liver disease is a major worldwide health problem. Diagnosis and staging of chronic liver diseases is an important issue. In this paper, we propose a quantitative method of analyzing local morphological changes for accurate and practical computer-aided diagnosis of cirrhosis. Our method is based on sparse and low-rank matrix decomposition, since the matrix of the liver shapes can be decomposed into two parts: a low-rank matrix, which can be considered similar to that of a normal liver, and a sparse error term that represents the local deformation. Compared with the previous global morphological analysis strategy based on the statistical shape model (SSM), our proposed method improves the accuracy of both normal and abnormal classifications. We also propose using the norm of the sparse error term as a simple measure for classification as normal or abnormal. The experimental results of the proposed method are better than those of the state-of-the-art SSM-based methods.

DOI： 10.1587/transinf.2014EDP7180

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC  

This study addresses the classification problem of the HEp-2 cell using indirect immunofluorescent (IIF) image analysis, which can indicate the presence of autoimmune diseases by finding antibodies in the patient serum. Generally, the method used for IIF analysis remains subjective, and depends too heavily on the experience and expertise of the physician. Recently, studies have shown that it is possible to identify the cell patterns using IIF image analysis and machine learning techniques. However, it still has large gap in recognition rates to the physical experts' one. This paper explores an approach in which the discriminative features of HEp-2 cell images in IIF are extracted and then, the patterns of the HEp-2 cell are identified using machine learning techniques. Motivated by the progress in the research field of computer vision, as a result of which small local pixel pattern distributions can now be highly discriminative, the proposed strategy employs a parametric probability process to model local image patches (textons: microstructures in the cell image) and extract the higher-order statistics of the model parameters for the image description. The proposed strategy can adaptively characterize the microtexton space of HEp-2 cell images as a generative probability model, and discover the parameters that yield a better fitting of the training space, which would lead to a more discriminant representation for the cell image. The simple linear support vector machine is used for cell pattern identification because of its low computational cost, in particular for large-scale datasets. Experiments using the open HEp-2 cell dataset used in the ICIP2013 contest validate that the proposed strategy can achieve a much better performance than the widely used local binary pattern (LBP) histogram and its extensions, rotation invariant co-occurrence LBP, and pair wise rotation invariant co-occurrence LBP, and that the achieved recognition error rate is even very significantly below the observed intralaboratory variability.

DOI： 10.1109/TBME.2014.2320294

J-GLOBAL

researchmap

Language：English   Publishing type：Part of collection (book)   Publisher：Springer New York  

This chapter focuses on the development of novel image enhancement and robust edge detection methods for practical medical image processing. It is known that the popular transformation-domain shrinkage approach for image enhancement applies a fixed mathematical basis to transform all images to be processed for noise or artifact reduction. However it is not adaptable to processed images, and then easily leads to blurring in the enhanced images. On the other hand, the techniques that are commonly used for edge detection are known as gradient and Laplacian operators (or mask), and smoothed gradient masks are typically used for edge detection in noisy images. However, these methods share a common major drawback wherein the associated masks are always fixed irrespective of the noise level in the images. In this study, we propose a novel learning-based method to adaptively deduce the transforming basis or masks from the processing data for medical image enhancement and robust edge detection. By using independent component analysis (ICA), the proposed learning-based method can extract suitable basis functions or masks for image transformation for processing data, which are adaptable to both the processed image and related noise in the image. The efficiency of the proposed learning-based method for medical image enhancement and edge detection is demonstrated experimentally using positron emission tomography (PET) and magnetic resonance imaging (MRI) medical images.

DOI： 10.1007/978-1-4614-7245-2_13

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IFAC Secretariat  

This study addresses the classification problem of HEp-2 cell using indirect immunofluores-cent (IIF) image analysis, which can indicate the presence of autoimmune diseases by searching for antibodies in the patient serum. Generally, IIF analysis remains a subjective method, which depends too heavily on the experience and expertise of the physician. Recently, some studies show that it is possible to identify the cell patterns using IIF by image analysis and machine learning techniques. However, it still has large gap between automatic recognition and the physical experts' decision. This paper explores the discriminative feature extraction of HEp-2 cell images in IIF, and then identifies the patterns of HEp-2 cell using machine learning techniques. Motivated by the research progress on computer vision that small local pixel pattern distributions can be highly discriminative, the proposed strategy employs a parametric probability process to model the local image patches (Textons: micro structure in the cell image), and extract the higher-order statistics (also called Fisher-Vector) to the model parameters for the image description. The proposed strategy can adaptively characterize the micro-Texton space of HEp-2 cell images as the generative probability model, and learn the parameters for better fitting the training space, which would lead to more discriminant representation for the cell image. The simple linear support vector machine is combined for cell pattern identification due to its low computational cost especially for large-scale dataset. Experiments on the released HEp-2 cell dataset of ICIP2013 competition validate that the proposed strategy can achieve much better performance than the popular used local binary pattern (LBP) image descriptor, and the achieved recognition error rate is even greatly below the observed intra-laboratory variability.

DOI： 10.3182/20140824-6-za-1003.01135

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：Springer Verlag  

This chapter concentrates the problem of recovery a high-resolution (HR) image from a single low-resolution input image. Recent research proposed to deal with the image super-resolution problem with sparse coding, which is based on the well reconstruction of any local image patch by a sparse linear combination of an appropriately chosen over-complete dictionary. Therein the chosen LR (Low-resolution) and HR (High-resolution) dictionaries have to be exactly corresponding for well reconstructing the local image patterns. However, the conventional sparse coding based image super-resolution usually achieves a global dictionary D=[D l
D h ] by jointly training the concatenated LR and HR local image patches, and then reconstruct the LR and HR image as a linear combination of the separated D l and D h . This strategy only can achieve the global minimum reconstructing error of LR and HR local patches, and usually cannot obtain the exactly corresponding LR and HR dictionaries. In addition, the accurate coefficients for reconstructing the HR image patch using HR dictionary D h are also unable to be estimated using only the LR input and the LR dictionary D l . Therefore, this paper proposes to firstly learn the HR dictionary D h from the features of the training HR local patches, and then propagates the HR dictionary to the LR one, called as HR2LR dictionary propagation, by mathematical proving and statistical analysis. The effectiveness of the proposed HR2LR dictionary propagation in sparse coding for super-resolution is demonstrated by comparison with the conventional super-resolution approaches such as sparse coding and interpolation. © 2014 Springer-Verlag Berlin Heidelberg.

DOI： 10.1007/978-3-642-54851-2_6

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE  

Magnetic resonance imaging can only acquire volume data with finite resolution due to various factors. In particular, the resolution in one direction (such as the slice direction) is much lower than others (such as the in-plane direction), yielding un-realistic visualizations. This study explores to reconstruct MRI isotropic resolution volumes from three orthogonal scans. This proposed super- resolution reconstruction is formulated as a maximum a posterior (MAP) problem, which relies on the generation model of the acquired scans from the unknown high-resolution volumes. Generally, the deviation ensemble of the reconstructed high-resolution (HR) volume from the available LR ones in the MAP is represented as a Gaussian distribution, which usually results in some noise and artifacts in the reconstructed HR volume. Therefore, this paper investigates a robust super-resolution by formulating the deviation set as a Laplace distribution, which assumes sparsity in the deviation ensemble based on the possible insight of the appeared large values only around some unexpected regions. In addition, in order to achieve reliable HR MRI volume, we integrates the priors such as bilateral total variation (BTV) and non-local mean (NLM) into the proposed MAP framework for suppressing artifacts and enriching visual detail. We validate the proposed robust SR strategy using MRI mouse data with high-definition resolution in two direction and low-resolution in one direction, which are imaged in three orthogonal scans: axial, coronal and sagittal planes. Experiments verifies that the proposed strategy can achieve much better HR MRI volumes than the conventional MAP method even with very high-magnification factor: 10. © 2014 SPIE.

DOI： 10.1117/12.2043361

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IOS PRESS  

Automatic tumor enhancement and detection has an essential role for the computer-aided diagnosis of liver tumor in CT volume data. This paper proposes a novel tumor enhancement strategy by extracting a tumor saliency map, which represents the uncommon or tumor tissue compared to the liver and vessel ones in CT volumes. The saliency map can be constructed by exploring the existing probability of tumor in any voxel. However, the tumor prototypes in a test liver volume from a specific patient or common tumor prototypes are extremely difficult to achieve due to requirement of full-searching and large variation of tumor tissues in different liver volumes. Therefore, this paper investigates a tumor-training-data free strategy by only constructing the common healthy liver and vessel prototypes, which can be extracted from any slice of a liver volume, and then applies a non-parametric Bayesian framework for calculating the existing probability of liver or vessel. Finally, the existing probability of tumor can be deduced from that of liver or vessel. The advantages of our proposed strategy mainly include three aspects: (1) it only needs to construct the prototypes of common tissue such as liver or vessel region, which are easily obtained in any liver volume; (2) it proposes an adaptive non-parametric framework for tumor enhancement, which does not need to learn a common classification model for all liver volumes; (3) dispensable to remove the other different structure such as vessel in liver volume as a pre-processing step. Experiments validate that the proposed Bayesian-based saliency model for liver tumor enhancement can perform much better than the conventional approaches such as EM, EM/MPM tumor segmentation methods.

DOI： 10.3233/978-1-61499-405-3-357

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：ELSEVIER SCIENCE BV  

This study addresses the classification problem of HEp-2 cell using indirect immunofluorescent (IIF) image analysis, which can indicate the presence of autoimmune diseases by searching for antibodies in the patient serum. Generally, IIF analysis remains a subjective method, which depends too heavily on the experience and expertise of the physician. Recently, some studies show that it is possible to identify the cell patterns using IIF by image analysis and machine learning techniques. However, it still has large gap between automatic recognition and the physical experts' decision. This paper explores the discriminative feature extraction of HEp-2 cell images in IIF, and then identifies the patterns of HEp-2 cell using machine learning techniques. Motivated by the research progress on computer vision that small local pixel pattern distributions can be highly discriminative, the proposed strategy employs a parametric probability process to model the local image patches (Textons: micro structure in the cell image), and extract the higher-order statistics (also called Fisher-Vector) to the model parameters for the image description. The proposed strategy can adaptively characterize the micro-Texton space of HEp-2 cell images as the generative probability model, and learn the parameters for better fitting the training space, which would lead to more discriminant representation for the cell image. The simple linear support vector machine is combined for cell pattern identification due to its low computational cost especially for large-scale dataset. Experiments on the released HEp-2 cell dataset of ICIP2013 competition validate that the proposed strategy can achieve much better performance than the popular used local binary pattern (LBP) image descriptor, and the achieved recognition error rate is even greatly below the observed intra-laboratory variability.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IOS PRESS  

A framework is proposed for automatic liver segmentation from CT volumes using probabilistic atlases and template matching techniques. Probabilistic atlases of human anatomy have been widely used for organ segmentation, which is used as a prior probability in a Bayes framework. The challenge is how to register the atlas to the patient volume. In this paper, we propose a template matching based technique for probabilistic atlas based organ segmentation. In our proposed method, we first find a Region of Interest (ROI) of the organ, which is based on human anatomic structure, and then the probabilistic atlas is used as a template to find the organ in the ROI by the use of template matching.

DOI： 10.3233/978-1-61499-405-3-412

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：HINDAWI PUBLISHING CORPORATION  

Face hallucination is one of learning-based super resolution techniques, which is focused on resolution enhancement of facial images. Though face hallucination is a powerful and useful technique, some detailed high-frequency components cannot be recovered. It also needs accurate alignment between training samples. In this paper, we propose a high-frequency compensation framework based on residual images for face hallucination method in order to improve the reconstruction performance. The basic idea of proposed framework is to reconstruct or estimate a residual image, which can be used to compensate the high-frequency components of the reconstructed high-resolution image. Three approaches based on our proposed framework are proposed. We also propose a patch-based alignment-free face hallucination. In the patch-based face hallucination, we first segment facial images into overlapping patches and construct training patch pairs. For an input low-resolution (LR) image, the overlapping patches are also used to obtain the corresponding high-resolution (HR) patches by face hallucination. The whole HR image can then be reconstructed by combining all of the HR patches. Experimental results show that the high-resolution images obtained using our proposed approaches can improve the quality of those obtained by conventional face hallucination method even if the training data set is unaligned.

DOI： 10.1155/2014/903160

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE-INT SOC OPTICAL ENGINEERING  

Magnetic resonance imaging can only acquire volume data with finite resolution due to various factors. In particular, the resolution in one direction (such as the slice direction) is much lower than others (such as the in-plane direction), yielding un-realistic visualizations. This study explores to reconstruct MRI isotropic resolution volumes from three orthogonal scans. This proposed super-resolution reconstruction is formulated as a maximum a posterior (MAP) problem, which relies on the generation model of the acquired scans from the unknown high-resolution volumes. Generally, the deviation ensemble of the reconstructed high-resolution (HR) volume from the available LR ones in the MAP is represented as a Gaussian distribution, which usually results in some noise and artifacts in the reconstructed HR volume. Therefore, this paper investigates a robust super-resolution by formulating the deviation set as a Laplace distribution, which assumes sparsity in the deviation ensemble based on the possible insight of the appeared large values only around some unexpected regions. In addition, in order to achieve reliable HR MRI volume, we integrates the priors such as bilateral total variation (B TV) and non-local mean (NLM) into the proposed MAP framework for suppressing artifacts and enriching visual detail. We validate the proposed robust SR strategy using MRI mouse data with high-definition resolution in two direction and low-resolution in one direction, which are imaged in three orthogonal scans: axial, coronal and sagittal planes. Experiments verifies that the proposed strategy can achieve much better HR MRI volumes than the conventional MAP method even with very high-magnification factor: 10.

DOI： 10.1117/12.2043361

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

Cirrhosis liver is a terrible disease which is threatening our lives. Meanwhile, cirrhosis will cause significant hepatic morphological changes. While it is well known that the livers from different subjects have similar global shape structure which means liver shape ensemble should be low-rank. However the deformation which caused by cirrhosis can be considered as sparse compared with the whole liver. Therefore, in this study, we proposed to apply spare and low-rank matrix decomposition to partition the local deformation part (sparse error matrix E) from the global similar structure (low-rank matrix A) using the input liver shape D, which is the landmark coordinates of liver shapes and already have been aligned by the current rigid registration methods firstly. And then sparse matrix E is used for diagnosis. In common sense, the normal liver should have less local deformation than that of abnormal liver, which means that the norm of sparse matrix E for normal liver is smaller than the norm for abnormal one. Thus, we can simply use a threshold classify normal and abnormal livers using the norm of E for these two categories. The proposed method is evaluated by a liver database which includes 30 normal livers and 30 abnormal livers. The experimental results of proposed method is better than those of state of the art statistical shape model(SSM) based methods.

DOI： 10.1109/ICPR.2014.579

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

Recent year, with the increasing of unhealthy diets which will threaten people's life due to the various resulted risks such as heart stroke, liver trouble and so on, the maintaining for healthy life has attracted much attention and then how to manage the dietary life is becoming more and more important. In this research, we aim to construct an auto-recognition system of food images and keep the daily food-log records which will contribute to manage dietary life. With the easily available food images taken by mobile phone, it prospects to give the insight about the daily dietary of users with our constructed food recognition system. In order to achieve the acceptable recognition performance of the food images, we propose to apply a sparse model for coding local descriptors extracted from the food images and various pooling methods for aggregating the xoded descriptors. Sparse coding: an extension of vector quantization for local descriptors, which is popularly used in Bag-of-Features (BoF) for image representation, can reconstruct the local descriptors more effective, and then obtain more discriminated feature for food image representation. However, in order to emphasize the strongest activated pattern, the widely applied aggregation strategy of the sparse coded vector is only to retain the maximum coefficient in all (named as Max-pooling), which would completely ignore the frequency: an important signature for identifying different types of images, of the activated patterns. Therefore, we explore a hybrid aggregation strategy named as top-ranked average pooling (TRAP), which integrates not only the maximum activated magnitude but also the stronger activated number for image representation. Experiments validate that the proposed hybrid aggregation strategy combined with sparse model can greatly improve the recognition rates compared with the conventional BOF model and the state-of-the-art methods on two databases: our constructed RFID and the public PFID.

DOI： 10.1109/ICPR.2014.265

researchmap

Language：Japanese   Publisher：(一社)日本医用画像工学会  

高精細医用画像から構築された人体臓器統計形状モデルは各臓器の個体差バリエーションを記述したモデルであり、計算解剖学をはじめ診断や治療支援への応用が期待できる。各臓器の3次元形状表現の有効な手法として球面調和関数を用いた臓器形状ベクトル表現が挙げられるが、臓器間における球面調和関数の次元数は異なるため、各臓器間の統計形状モデル構築に対する次元数特定やその差異に対する統計形状モデル性能評価が確立されていない。故に球面調和関数による臓器統計形状モデルの支援診断へ応用は未着手である。本研究では、球面調和関数を用いた信頼性の高い統計形状モデル構築を目指し、球面調和関数による人体の各臓器3次元形状表現の有効性と統計形状モデル構築、性能評価を行った。本研究では、球面調和関数を用いた人体形状の3次元形状表現を用いることで、効率的な統計形状モデル構築が可能であることを示す。(著者抄録)

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

We propose a multilinear independent component analysis (ICA) framework called generalized N-dimensional ICA (GND-ICA) by extending the conventional linear ICA based on the multilinear algebra. Unlike the linear ICA that only treats one-dimensional data, the proposed GND-ICA treats N-dimensional data as a tensor without any preprocess of data vectorization. We furthermore introduce two types of GND-ICA solutions and analyze their efficiency and effectiveness. As an application, the GND-ICA can be used for multiple feature fusion and representation for color image classification. Many features extracted from a given image are constructed as a tensor. The feature tensor can be effective represented by GND-ICA. Compared with the conventional linear subspace learning methods, GND-ICA is capable of obtaining more distinctive representation for color image classification. © 2012 Elsevier B.V.

DOI： 10.1016/j.neucom.2012.09.020

Scopus

researchmap

Language：English   Publisher：Japanese Society for Medical and Biological Engineering  

DOI： 10.11239/jsmbe.51.M-85

CiNii Article

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

The trait anxiety degree is a significant standard to measure the psychological status, and the measurement (score) of trait anxiety degree is generally obtained by a very complex text questionnaire, which usually takes large amount of time and is subjectively various according to environmental condition. On the other hand, the researches in psychological field have proven that personality recognition of different facial emotions is strongly related to the degree of trait anxiety. In this work, we propose to automatically predict the trait anxiety score using the recognition rates of different facial emotions. In order to select compact and discriminant features, we investigate a correlation-based feature selection strategy in both raw data and PCA transformed space. Experimental results show that our proposed strategy can achieve reasonable trait anxiety score, which, also can validates the reliable relation between recognition rates of facial emotion and trait anxiety score. © 2013 IEEE.

DOI： 10.1109/ICACI.2013.6748515

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

This study addresses the problem of generating a high-resolution (HR) MRI volume from a single low-resolution (LR) MRI input volume. Recent researches have proved that sparse coding can be successfully applied for single-frame super-resolution for natural images, which is based on good reconstruction of any local image patch with a sparse linear combination of atoms taken from an appropriate over-complete dictionary. This study adapts the basic idea of sparse code-based super-resolution (SCSR) for MRI volume data, and then improves the dictionary learning strategy in the conventional SCSR for achieving the precise sparse representation of HR volume patches. In the proposed MRI super-resolution strategy, we only learn the dictionary of the HR MRI volume patches with sparse coding algorithm, and then propagate the HR dictionary to the LR dictionary by mathematical analysis for calculating the sparse representation (coefficients) of any LR local input volume patch. The unknown corresponding HR volume patch can be reconstructed with the sparse coefficients from the LR volume patch and the corresponding HR dictionary. We validate that the proposed SCSR strategy through dictionary propagation can recover much clearer and more accurate HR MRI volume than the conventional interpolated methods. © 2013 SPIE.

DOI： 10.1117/12.2008145

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：HINDAWI LTD  

It is widely known that morphological changes of the liver and the spleen occur during the clinical course of chronic liver diseases. In this paper, we proposed a morphological analysis method based on statistical shape models (SSMs) of the liver and spleen for computer-aided diagnosis and quantification of the chronic liver. We constructed not only the liver SSM but also the spleen SSM and a joint SSM of the liver and the spleen for a morphologic analysis of the cirrhotic liver in CT images. The effective modes are selected based on both its accumulation contribution rate and its correlation with doctor's opinions (stage labels). We then learn a mapping function between the selected mode and the stage of chronic liver. The mapping function was used for diagnosis and staging of chronic liver diseases.

DOI： 10.1155/2013/264809

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

It is widely known that morphological changes of the liver and the spleen occur during the clinical course of chronic liver diseases. In this paper, we show a preliminary study on quantifying stage progress of cirrhotic liver based on statistical shape models (SSMs). We constructed not only the liver SSM, but also the spleen SSM and a joint SSM of the liver and the spleen for a morphologic analysis of the cirrhotic liver in CT images. The effective modes are selected based on both its accumulation contribution rate and its correlation with doctor's opinions (stage labels). Both normal and abnormal livers including temporal sequence data are projected to the subspace with the selected modes. Most of the normal data are located in the center, while abnormal data are scattered around the normal data. If the disease progresses, data will move outside. The distance to the center can be used to quantify the stage progress of the cirrhotic liver.

DOI： 10.1109/BMEI.2013.6747054

researchmap

Language：Japanese   Publisher：The Institute of Electrical Engineers of Japan  

In recent years, there are increasing interests in statistical shape modeling of human anatomy. The statistical shape model can capture the morphological variations of human anatomy. Since the liver cirrhosis will cause significant morphological changes, the authors propose a computer-aided diagnosis method for liver cirrhosis based on statistical shape models. In the proposed method, the authors first construct a statistical shape model of the liver using 50 clinical CT datasets (25 sets of normal data and 25 sets of abnormal data). The authors apply marching cubes algorithm to convert the segmented liver volume to a triangulated mesh surface which containing 1000 vertex points. The coordinates of these vertex points are used to represent 3D liver shape as a shape vector. After normalization and correspondence finding between all datasets, Principal Component Analysis (PCA) is employed to find the principal variation modes of shape vectors. Then the authors propose a mode selection method based on class variations between the normal class and abnormal class. The authors found the top two modes of class variations are most effective for classification of normal and abnormal livers. The classification rate of abnormal livers and normal liver are 84% and 80%, respectively, by the use of a simple linear discriminant function.

DOI： 10.1541/ieejeiss.133.2037

CiNii Article

researchmap

researchmap

Language：Japanese   Publisher：The Japanese Society of Medical Imaging Technology  

Computer-assisted diagnosis/surgery systems have had a significant influence on the diagnosis and surgical treatment of hepatic diseases. In this paper, we propose a novel simulation system for liver surgical planning that combines image processing and computer vision techniques. The proposed system is composed of three modules: liver segmentation, vessel extraction, and visualization/interaction. We first segment the liver region from a CT volume using K-means clustering and geodesic active contour algorithms. We then extract the vessels using multiscale filters. The third visualization/interaction module visualizes the vessels and prepares a virtual environment for the user to perform surgical procedures on the liver image. First, the internal structure of the liver together with the vessels is shown to the physician. Virtual surgery is then started. A toggling option makes it easy to view the internal structure of the liver or make it opaque during surgery. This system is expected to be useful for treatment/surgical planning and may even serve as a guided surgery system.

DOI： 10.11409/mit.31.176

CiNii Article

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：WILEY-BLACKWELL  

Super-resolution is a process for obtaining high-quality, high-resolution images from one or a set of low-resolution images. The most practical methods for image super-resolution are reconstruction-based methods, which minimize the difference between observed low-resolution images and the estimate for high-resolution images. Therein, the interpolation step plays a key role in the estimated high-resolution image quality. Usually, the conventional bilinear or bicubic methods are used in reconstruction-based super-resolution. However, these conventional interpolations generally lead to blurring in edge regions and need more time for convergence in the reconstruction-based super-resolution method. Therefore, this paper proposes a gradient-based edge-preserving interpolation method, which can reduce not only artifact noise but also blurring near the edge regions in the estimated high-resolution image. Furthermore, our proposed interpolation method can also solve high-complexity, time-consuming problems in the recently developed new edge-directed interpolation, which usually can achieve sharp edges in the high-resolution reconstructed image. Experiments confirm that our proposed interpolation method for image super-resolution is more effective than the conventional interpolation methods. (c) 2012 Wiley Periodicals, Inc. Electron Comm Jpn, 96(1): 4350, 2013; Published online in Wiley Online Library (wileyonlinelibrary.com). DOI 10.1002/ecj.11413

DOI： 10.1002/ecj.11413

researchmap

Publisher：IEEE  

DOI： 10.1109/SII.2013.6776758

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Recently, with the increasing of unhealthy diets and the attracted attention for healthy life, how to manage the dietary life is becoming more and more important. In this paper, we aim to construct a food-log system, which can auto-recognize the menu contents from food image taken by mobile phone. In order to increase recognition rate of food image, this research explores the typical codebook-based mode-Bag-of Feature (BOF), and the improved sparse-coding (Sc) based one for image representation, which is one of the main factors for affecting on food-log system. Furthermore, instead of pooling all the coded local features globally, the Spatial Pyramid Matching (SPM) strategy is adopted to integrated them from various spatial scale sub-regions. The experimental results validate that the Sc-based codebook model can achieve much better performance than the typically Bag-of-Feature on dish recognition.

DOI： 10.1109/SII.2013.6776730

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer Verlag  

The paper addresses the issues of compact representation of medical volumes and effective mode selection. Medical volumes are unlike traditional two dimensional images in pattern recognition which in spatial domain has three dimensions. Meanwhile it often meets small sample problem. Thus common compact representation methods, such as PCA, do not fit medical volumes. Because the number of eigenvector is so little that it will lose much useful information. In previous work, we proposed a Generalized N-dimensional principal component analysis (GND-PCA) for reconstruction of medical volumes with only few samples. The core tensor of GND-PCA can keep most useful information. However, making diagnosis using the core tensor is difficult due to most modes very general for all samples. It will affect finally diagnosis. To resolve this problem, Adaboost is used as classifier in diagnosis because it can choose distinctive mode according to its definition. The proposed method was evaluated using a medical volume database. In our experiment, we compare Adaboost with SVM and KNN. The classification accuracy of Adaboost is slightly better than that of SVM and KNN,meanwhile time consuming of classification of Adaboost is greatly less than those of other two classification methods. © Springer International Publishing Switzerland 2013.

DOI： 10.1007/978-3-319-03731-8_58

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE-INT SOC OPTICAL ENGINEERING  

This study addresses the problem of generating a high-resolution (HR) MRI volume from a single low-resolution (LR) MRI input volume. Recent researches have proved that sparse coding can be successfully applied for single-frame super-resolution for natural images, which is based on good reconstruction of any local image patch with a sparse linear combination of atoms taken from an appropriate over-complete dictionary. This study adapts the basic idea of sparse code-based super-resolution (SCSR) for MRI volume data, and then improves the dictionary learning strategy in the conventional SCSR for achieving the precise sparse representation of HR volume patches. In the proposed MRI super-resolution strategy, we only learn the dictionary of the HR MRI volume patches with sparse coding algorithm, and then propagate the HR dictionary to the LR dictionary by mathematical analysis for calculating the sparse representation (coefficients) of any LR local input volume patch. The unknown corresponding HR volume patch can be reconstructed with the sparse coefficients from the LR volume patch and the corresponding HR dictionary. We validate that the proposed SCSR strategy through dictionary propagation can recover much clearer and more accurate HR MRI volume than the conventional interpolated methods.

DOI： 10.1117/12.2008145

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：Springer  

Recently a growing interest has been seen in single-frame super-resolution techniques, which are known as example-based or learning based super-resolution techniques. Face Hallucination is one of such techniques, which is focused on resolution enhancement of facial images. Though face hallucination is a powerful and useful technique, some detailed high-frequency components cannot be recovered. In this paper, we propose a high-frequency compensation framework based on residual images for face hallucination method in order to improve the reconstruction performance. The basic idea of proposed framework is to reconstruct or estimate a residual image, which can be used to compensate the high-frequency components of the reconstructed high-resolution image. Three approaches based on our proposed framework are proposed. Experimental results show that the high-resolution images obtained using our proposed approaches can improve the quality of those obtained by conventional face hallucination method. © 2013 Springer-Verlag Berlin Heidelberg.

DOI： 10.1007/978-3-642-39065-4-75

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper explores the potential of applying shape analysis to classify normal/cirrhotic liver and in addition estimate the severity of abnormal cases. Conventional Computer-Aided Diagnosis (CAD) systems are developed for automatically providing a binary output as a second opinion to assist radiologists to draw conclusions about the condition of the pathology (normal or abnormal). After the disease is diagnosed, grasping the proceeding stage of the abnormal degree is essential for adopting the appropriate strength of treatment. However, none of existing CAD system is well established for such a challenging task. Liver cirrhosis has an important feature: morphological changes of the liver and the spleen occur during the clinical course of liver cirrhosis. In this study we constructed liver, spleen and their joint Statistical Shape Models (SSMs) to quantitatively assess the global shape variation and selected several modes from the SSMs. Then we learnt a mapping function between coefficients of selected modes and the ground truth staging label by Support Vector Regression (SVR). Using this mapping function, the proceeding stage of new input data can be estimated. Experimental results have validated the potential of our method on assisting the cirrhosis diagnosis.

DOI： 10.1109/ICIP.2013.6738730

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

Cirrhosis will cause significant morphological changes on both liver and spleen. In this paper, we constructed not only the liver statistical shape models (SSM), but also the spleen SSM and a joint SSM of the liver and the spleen for a morphologic analysis of the cirrhotic liver in CT images. We also proposed a mode selection method based on both its accumulation contribution rate and its correlation with doctor's opinions (labels). The classification performance for normal and abnormal livers is significantly improved by our proposed method. The classification accuracies for normal and cirrhotic livers are 88% and 90%, respectively.

DOI： 10.1007/978-3-642-39094-4_56

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/EMBC.2013.6610012

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Recent year, with the increasing of unhealthy diets which will threaten people's life due to the various resulted risks such as heart stroke, liver trouble and so on, the remain for healthy life has attracted much attention and then how to manage the dietary life is becoming more and more important. In this research, we aim to construct a auto-recognition system of food images and keep the daily food-log records which will contribute to manage dietary life. With the easily available food images taken by mobile phone, it prospects to give the insight about the daily dietary of users with our constructed food recognition system. In order to achieve the acceptable recognition performance of the food images, we propose to apply a sparse model for coding a local descriptor extracted from the food images. Sparse coding: an extension of vector quantization for local descriptors, which is popularly used in Bag-of-Features (BoF) for image representation in generic object recognition, can represent the local descriptors more efficient, and then abtain more discriminant feature for food image representation. Moreover, in order to introduce spatial information, a hierarchic spatial structure is explored to extract the feature based sparse model. Experiments validate that the proposed strategy can greatly improve the recognition rates compared with the conventional BOF model on two databases: our constructed RFID and the public PFID.

DOI： 10.1109/BMEI.2013.6747060

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper proposes a new image representation for texture categorization, which is based on extension of local binary patterns (LBP). As we know LBP can achieve effective description ability with appearance invariance and adaptability of patch matching based methods. However, LBP only thresholds the differential values between neighborhood pixels and the focused one to 0 or 1, which is very sensitive to noise existing in the processed image. This study extends LBP to local ternary patterns (LTP), which considers the differential values between neighborhood pixels and the focused one as negative or positive stimulus if the absolute differential value is large; otherwise no stimulus (set as 0). With the ternary values of all neighbored pixels, we can achieve a pattern index for each local patch, and then extract the pattern histogram for image representation. Experiments on two texture datasets: Brodats32 and KTH TIPS2-a validate that the robust LTP can achieve much better performances than the conventional LBP and the state-of-the-art methods.

DOI： 10.1109/BMEI.2013.6747059

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Face Hallucination, one of the most famous learning-based image super-resolution techniques for facial images, can reconstruct a high-resolution image using only one low-resolution image. However, some detailed high-frequency components of the reconstructed image cannot be recovered using this method. In addition, the available LR images are sometimes blurred because of object movement or hardware problems. In this study, we proposed a high-frequency compensated face hallucination method with total variation constraint for HR image recovery. The proposed method is divided into two main processes: 1) Deblurring the LR input with total variation constraint; 2) Super-resolution with the proposed high-frequency compensated face hallucination. Experimental results show that the highresolution images obtained by our proposed approach are much better than those obtained by conventional methods.

DOI： 10.1109/BMEI.2013.6747056

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Semantic understanding of images remains an important research challenge in machine intelligence and statistical learning. It mainly includes two steps: feature extraction and classification. This study mainly focuses scene image recognition, where color information plays an important role. The conventional color representation of images mainly includes color distribution (histogram) and its statistical information based on uniformed quantization color bin. However, for a specific recognition application such several scene types, some quantized colors maybe never appear in any scene image, and at the same time the detail variation in other quantized colors include much discriminative features. Therefore, this study proposes to characterize the color information of scene images using a leaning strategy for producing adaptive color level, and extract the histogram of the learned color levels for image representation. With the proposed strategy, the compact (learned) color levels can represent the image in our application more faithful than the uniform quantized conventional RGB color. Experimental results show that the recognition rate with our proposed methods can be greatly improved compared to the conventional color histogram.

DOI： 10.1109/BMEI.2013.6747055

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

The assessment of the treated margin with locoregional therapy (LT), for hepatocellular carcinoma (HCC), is the common method for predicting HCC recurrence in most hospital. However, tumors sometimes cannot be removed clearly with LT in limited conditions. The therapeutic efficiency of HCC is often evaluated by comparing 2D fusion images of computed tomography (CT) or magnetic resonance imaging (MRI) between the preoperation and the postoperation. However, judgment about whether the tumors exist in the treated margin after LT by using 2D slices sometimes is difficult. It is desirable to develop a suitable image registration algorithm to automatically align the two volumes in order to transform the treated margin of the postoperative volume to the tumor of the preoperative volume to assess the therapeutic efficiency after treatment of HCC. With taking these into consideration, this paper proposed an automatic 3D fusion imaging approach for medical image by using the nonrigid registration method that aligning an ablative marginthat is the treated margin after LT, onto the locations of HCC. In our registration algorithm, a rigid global transformation combined with localized B-spline is used to estimate the significant nonrigid motions of the liver between before and after LT. Our proposed approach can ensure the feasibility, the accuracy and the efficacy to assess the treated margin for HCC. Furthermore, this method can be adapted to register multimodality medical images. We demonstrate the effectiveness of our proposed method by comparing the difference criterions of fusion evaluation on medical images. The results clearly indicate that our method extremely useful in the evaluation of the treated margin, in addition, it remain the motion and local deformation of the volume.

DOI： 10.1109/BMEI.2013.6747052

researchmap

Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

DOI： 10.1109/BMEI.2013.6747049

researchmap

Publishing type：Research paper (scientific journal)  

DOI： 10.1155/2013/264809

researchmap

Language：Japanese   Publisher：The Institute of Electrical Engineers of Japan  

This paper presents a new method for single-frame Super-Resolution (SR), by combining Example-based SR and neighbor embedding based SR (NE-based SR). Example-based SR attempts to generate High-Resolution (HR) image through estimating the High-Frequency (HF) components that are lost in the input Low-Resolution (LR) image. This method usually can achieve acceptable HR images if enough amounts of similar training samples are prepared. However, the HF component is approximated by only one training sample, which easily produces noise and artifacts. On the other hand, NE-based SR recovers HR image using manifold learning - Locally Linear Embedding, which represents any LR input and its corresponding HR one by a weighted linear combination of several training patches. The NE-based SR need to prepare large-scale training database with both intensity and structure variation, which will lead to high computation. This study combines these two methods to only estimate the HF components using several training samples. Moreover, we extend the proposed method to a fast version by processing only the patches with large variance. Experimental results show that the reconstructed HR images by our proposed approach are much better than those by conventional methods and interpolation techniques, and at the same time the computation is much faster.

DOI： 10.1541/ieejeiss.132.1768

Scopus

CiNii Article

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

Medical images must be stored and translated in a secure way. This paper proposes a multiple watermarks technique based on medical image features to serve these purposes. A part of sign sequence of DWT and DFT coefficients are used as feature vector for enhancing the robustness against rotation, scaling, translation and cropping attacks. In contrast to conventional approaches, our method can strengthen the fidelity of images through considering local image characteristics, without sacrificing the quality of medical images. We describe how to exact the feature vector of medical image, and then, embed and extract the multiple watermarks. The simulation results demonstrate that the watermarking scheme has strong robustness, and can embed much more data compared with the existing watermarking techniques.

DOI： 10.4156/jdcta.vol6.issue10.28

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

In this paper, we propose N-Dimensional (ND) Tensor Supervised Neighborhood Embedding (ND TSNE) for discriminant feature representation, which is used for view-based object recognition. ND TSNE uses a general Nth order tensor discriminant and neighborhood-embedding analysis approach for object representation. The benefits of ND TSNE include: (1) a natural way of representing data without losing 'structure information, i.e., the information about the relative positions of pixels or regions; (2) a reduction in the small sample size problem, which occurs in conventional supervised learning because the number of training samples is much less than the dimensionality of the feature space; (3) preserving a neighborhood structure in tensor feature space for object recognition and a good convergence properly in training procedure. With Tensor-subspace features, the random forests is used as a multi-way classifier for object recognition, which is much easier for training and testing compared with multi-way SVM. We demonstrate the performance advantages of our proposed approach over existing techniques using experiments on the COIL-100 and the ETH-80 datasets.

DOI： 10.1587/transinf.E95.D.835

J-GLOBAL

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC  

In this paper, we propose to represent an image as a local descriptor tensor and use a multilinear supervised neighborhood embedding (MSNE) for discriminant feature extraction, which is able to be used for subject or scene recognition. The contributions of this paper include: 1) a novel feature extraction approach denoted as the histogram of orientation weighted with a normalized gradient (NHOG) for local region representation, which is robust to large illumination variation in an image; 2) an image representation framework denoted as the local descriptor tensor, which can effectively combine a moderate amount of local features together for image representation and be more efficient than the popular existing bag-of-feature model; and 3) an MSNE analysis algorithm, which can directly deal with the local descriptor tensor for extracting discriminant and compact features and, at the same time, preserve neighborhood structure in tensor-feature space for subject/scene recognition. We demonstrate the performance advantages of our proposed approach over existing techniques on different types of benchmark database such as a scene data set (i.e., OT8), face data sets (i.e., YALE and PIE), and view-based object data sets (COIL-100 and ETH-80).

DOI： 10.1109/TIP.2011.2168417

J-GLOBAL

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Example-Based Super-Resolution is a learning-based technique that attempts to recover high-resolution (HR) image according to the corresponding relation in a set of training low-resolution (LR) and high-resolution image pairs prepared in advance. The conventional learning-based method for image super-resolution usually cannot achieve the high-frequency components accurately, which are lost in the input LR image, for recovering the HR image, since it only estimates the lost information using one most similar training LR patch to the input patch, and its corresponding HR pair. Therefore, we propose to use a manifold learning method- Locally Linear Embedding (LLE) for reconstructing the input LR patch with a linear weight summation of its several most similar training LR patches, and then can recover HR patch using the same linear summation of the corresponding training HR patches. Furthermore, in order to solve the expensive computational problem in the conventional exampled-based learning method, only the patches with larger variance, which means with high-frequency components, are selected for super-resolution procedures. Finally, Experimental results show that the recovered high-resolution images by our proposed approach are much better than those by conventional method and interpolation techniques.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Recently, with the increasing of unhealthy diets and the attracted attention for healthy life, how to manage the dietary life is becoming more and more important. In this paper, we aim to construct a system, which can auto-recognize the menu contents from food image taken by mobile phone. As we know that the viewpoints can be varied in any direction when taking food images, and then, rotation-robust features for image representation are very important. Therefore, in this paper, we propose to extract rotation invariant features using circle-segmentation called SPIN for food recognition, and construct a Food-Log system, which records the contents of food menu, calories and nutritional value for management of the dietary life.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper presents a method for robust watermarking of medical image using DFT. A feature vector of the medical image is utilized to enhance the robustness against rotation, scaling, and translation. The proposed algorithm utilizes the medical image's feature, Hash function, the third party authentication. We describe how to obtain the feature vector of medical image and embed and extract the watermarking. Simulation results demonstrate the proposed algorithm's robustness against common and geometrical attacks.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Medical images must be stored in a secure way to preserve unauthorized disclosure of patient data. This paper proposes an algorithm of robust multiple watermarks to serve these purposes using DWT and DFT. A part of sign sequence of DWT and DFT coefficients are used as feature vector for enhancing the robustness against rotation attacks, scaling attacks, translation attacks and cropping attacks. Moreover, the content of medical image remains unchanged with our proposed algorithm which is one kind of the zero-watermarking technology. We describe how to extract the feature vector of medical image, and then, embed and extract the multiple watermarks. The results of experiment indicate that the watermark scheme has strong robustness, and can embed much more data compared with the existing watermarking techniques.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Linear coding is used for finding succinct representations of data sets. It also discover basis functions that capture higher-level features in the data. However, finding linear codes for multi-dimensional data remains a very difficult computational problem. Motivated by the work of linear image coding and multilinear algebra, we propose a linear tensor coding algorithm (LTC), which is applied to represent multi-dimensional data succinctly by a linear combination of tensor-formed bases without data expansion. Each basis captures some specific variability. The coefficients of data, which are associated with the bases, can be applied for representation, compression and classification. When we applied LTC algorithm on the phantom data, experimental results illustrate that our algorithm not only produces localized bases but also preserve the information of the input data.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In medical imaging, the data resolution is usually insufficient for accurate diagnosis in clinical medicine. Especially in most case, the resolution in the slice direction (Z direction) is much lower than that of the in-plane resolution (XY direction). Therefore it is difficult to construct isotropic voxels, which is very important in 3-D visualization systems, such as surgical system. In this paper, we propose a method for improving resolution in the slice direction for medical volume images based on Principal Component Regression (PCR), which can be considered as one of the learning based super-resolution techniques. The experimental results verify the effectiveness of the proposed method by comparison with the conventional interpolation methods.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In computational anatomy, statistical shape model is used for quantitative evaluation of the variations of an organ shape. Since liver cirrhosis will cause significant hepatic morphological changes, we applied statistical shape model of the liver to capture the morphological changes and recognize whether a liver is normal or abnormal. In this paper, we propose an effective mode selection method to improve the classification accuracy. In addition to the conventional Accumulated Variance Contribution Rate (AVCR) based mode selection, we newly propose a Pearson correlation based mode selection method and combine them to select the effective modes. The coefficients of the selected modes (components) are used as features to recognize whether liver is normal or abnormal. The effectiveness of the proposed method is evaluated by the classification accuracy of normal and abnormal. Experimental results show that our proposed method is superior than conventional methods.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In clinical course of the chronic liver disease, the liver finally develops liver cirrhosis from the state of the chronic hepatitis. In the process, the patients are complicated with esophagogastric varices due to the portal hypertension and a hepatocellular carcinoma. It is very useful to be diagnosed using MDCT morphologically. A quantitative assessment has been tried from the CT images of the liver. A morphologic diagnosis of the chronic liver disease using the CT and a quantitative assessment are described in this article.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In medical imaging research, the three-dimensional (3-D) shape representation and analysis of anatomic structures using only a few parameters is an important issues, and can be applied to computer assisted diagnosis, surgical simulations, visualization, and many other medical applications. This paper proposes the representation of the shape surfaces of a simply connected 3-D object using spherical harmonic (spharm) functions, which can provide an approximate global description of the shape using only a few spharm parameters. Spharms are obtained from several partial differential equations used in physics, such as the Laplace, Helmholtz, and Schrodinger equations, in which spherical coordinates are used. In this study, we aim to develop a 3D representation of human anatomy, such as the liver and spleen, with fewer parameters by using spharm functions. Furthermore, the experimental results show that the shape of the liver and spleen can be accurately represented with only a few spharm parameters.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we propose to represent a face image as a local descriptor tensor and use a Multilinear Manifold Analysis (MMA) method for discriminant feature extraction, which is used for face recognition. The local descriptor tensor, which is a combination of the descriptor of local regions (K*K-pixel patch) in the image, can represent image more efficient than pixel-level intensity representation, and also than the popular Bag-Of-Feature (BOF) model, which approximately represents each local descriptor as a predefined visual word. Therefore it should be more effective in computational time than the BOF model. For extracting discriminant and compact features from the local descriptor tensor, we propose to use the proposed TMultilinear Manifold Analysis (MMA) algorithm, which has several benefits compared with conventional subspace learning methods such as PCA, ICA, LDA and so on: (1) a natural way of representing data without losing structure information, i.e., the information about the relative positions of pixels or regions; (2) a reduction in the small sample size problem which occurs in conventional supervised learning because the number of training samples is much less than the dimensionality of the feature space; (3) a neighborhood structure preserving in tensor feature space for face recognition and a good convergence property in training procedure. We validate our proposed algorithm on Benchmark database Yale and PIE, and experimental results show recognition rate with the proposed method can be greatly improved compared with conventional subspace analysis methods especially for small training sample number ....

DOI： 10.1007/978-3-642-34166-3_81

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In the field of medical image analysis, the three-dimensional (3-D) shape representation and modeling of anatomic structures using only a few parameters is an important issue, and can be applied to computer assisted diagnosis, surgical simulations, visualization, and many other medical applications. In this paper, we show that the 3D anatomical structure such as the liver can be represented by a few coefficients of spherical harmonic functions (SPHARM). We also propose to use SPHARM based shape representation for statistical shape modeling. Since the dimension of SPHARM based shape representation vector is much lower than the conventional shape representation using coordinates of surface points, our proposed method can be used for small number of training samples and enhance the computation cost.

DOI： 10.1109/SCIS-ISIS.2012.6505370

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Magnetic resonance imaging can only acquire volume data with finite resolution due to various factors. In particular, the resolution in the slice direction is much lower than that in the in-plane direction, yielding un-realistic visualizations. To solve this problem, interpolation techniques have conventionally been applied. However, classical interpolation techniques generally cause some artifact noise such as jaggedness and blurring in the edge regions. In this paper, we propose a new super-resolution framework for generating high-resolution data in the slice direction. In the proposed approach, we estimate the high-frequency component using a learning-based super-resolution technique with sparse representation and prove that the dictionary can be constructed using the in-plane frame as the input data without any other high-resolution data as training. Furthermore, we optimize estimated high-resolution data by adding a new regularization term with a non-local means algorithm. Experiments confirm that our proposed method is more effective than the conventional methods.

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/icpr/icpr2012.html#conf/icpr/IwamotoHSTXC12

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper presents a learning-based method called image super-resolution (SR) for generating a high-resolution (HR) image from a single low-resolution (LR) image. Recent research investigated the image SR problem using sparse coding, which is based on good reconstruction of any image local patch by a sparse linear combination of atoms from an overcomplete dictionary. However, sparse-coding-based SR (ScSR) generally takes a significant amount of computational time to compute an HR image. Further, it can yield only a global dictionary D = [D-h;D-l] by jointly training the concatenated HR and LR image local patches, which results in no accurate correspondence between the HR and LR dictionaries. Therefore, we propose the generation of an HR image using a linear combination of several anchor points (codes) for a local patch based on locality-constrained linear coding (LLC), which is a fast implementation of local coordinate coding (LCC). In the proposed LLC-based strategy, each local patch is represented by a weighted linear combination of its nearer codes in a predefined codebook, and the linear weights become its local coordinate coding. Experimental results show that the recovered HR images with our proposed approach can achieve comparable performance at a processing time much shorter than those of conventional methods.

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/icpr/icpr2012.html#conf/icpr/TaniguchiHISC12

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Recent years have seen an increasing interest in codebook-based model(bag-of-words BOW) for image representation, which includes the basic bag-of-words model and its improved version for local descriptor reconstruction with sparse coding (SC) and locality-constrained linear coding (LLC) etc.. Although the recent coding strategies in the BoW model can lead to prospect performance using large amounts of codes (codebooks) for image classification, it is usually computational expensive for obtaining the global image representation through calculating the similarities between each local descriptor and all codes. Therefore, this study proposes to represent a local descriptor with an adaptive code or its variation modes (adaptive subdomain) in a small set of codebooks. The proposed strategy can adaptively select one code to saliently representation, or adaptively select one sub-domain of a code for group sparse reconstruction of a local descriptor. Due to computational cost mainly on the similarity calculation between local descriptors and the predefined codebooks, our proposed strategy using small set of codebook can greatly reduce computational time, and in addition, shows prospect performances for image classification on an scene database, called OMRON scene dataset, and the benchmark data: 15 natural scene dataset.

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/icpr/icpr2012.html#conf/icpr/HanQC12

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper proposes a framework of tensor-based ICA method for N-dimensional data analysis, which is called generalized N-dimensional ICA (GND-ICA). The proposed GND-ICA is based on multilinear algebra that treats N-dimensional data as a tensor without any unfolding preprocess. As an application, the GND-ICA can be used for multiple feature fusion and representation for color image classification. Multiple features extracted from a given image are constructed as a tensor. The effective components for each feature can be selected simultaneously and combined by the GND-ICA. This can obtain the improved classification results in comparison with various conventional linear and multilinear subspace learning methods.

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/icpr/icpr2012.html#conf/icpr/AiDHC12

Language：Japanese   Publisher：The Institute of Electrical Engineers of Japan  

Super-resolution is a process for obtaining high quality and high resolution images from a set of or only one low-resolution image. The most practical one for image super-resolution is reconstruction-based method, which minimizes the difference between observed low-resolution images and the estimation for high resolution image. Therein, interpolation step plays a key role for the estimated high resolution image quality. Usually, the conventional bilinear or bicubic methods are used in the reconstruction-based super-resolution. However, these conventional interpolations generally lead to blurring on edge regions and need more time for convergence in reconstruction-based super-resolution method. Therefore, this paper propose a gradient based edge preserving interpolation method, which can reduce not only artifact noise but also blurring near edge regions in the estimated high resolution image. Furthermore, our proposed interpolation method can also solve large complexity and time-consuming problem in the recently developed New Edge-directed interpolation, which usually can achieve sharp edge in the high resolution reconstructed image. Experiments validate that our proposed interpolation method for image super-resolution is more effective than the conventional interpolation ones.

DOI： 10.1541/ieejeiss.131.1901

Scopus

CiNii Article

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

This paper addresses the problem of ordering the color SIFT descriptors in the independent component analysis for image classification. Component ordering is of great importance for image classification, since it is the foundation of feature selection. To select distinctive and compact independent components (IC) of the color SIFT descriptors, we propose two ordering approaches based on local variation, named as the localization-based IC ordering and the sparseness-based IC ordering. We evaluate the performance of proposed methods, the conventional IC selection method (global variation based components selection) and original color SIFT descriptors on object and scene databases, and obtain the following two main results. First, the proposed methods are able to obtain acceptable classification results in comparison with original color SIFT descriptors. Second, the highest classification rate can be obtained by using the global selection method in the scene database, while the local ordering methods give the best performance for the object database.

DOI： 10.1587/transinf.E94.D.1800

J-GLOBAL

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：ICIC INT  

Image category recognition is important to access visual information on the level of objects and scene types. In this paper, we develop a Supervised Nonlinear Neighborhood Embedding (SNNE) subspace algorithm of different visual features for object and scene recognition, which learns an adaptive nonlinear subspace by preserving the neighborhood structure of the visual feature space. In the proposed subspace algorithm, we combine the idea of nonlinear kernel mapping and preserving the neighborhood structure of the samples, so it can not only gain a perfect approximation of the nonlinear image manifold, but also enhance within-class neighborhood information. So, the proposed SNNE algorithm models the ensemble of visual features to a more discriminative space for category recognition, and at the same time, can effectively combine several visual features to improve recognition rate. The proposed method is evaluated by using the scene database (SIMPLicity) and object recognition database (Caltech). We confirm that the proposed method is much better than state-of-the-art methods only with simple visual features.

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：SPRINGER LONDON LTD  

In this paper, a robust edge detection method based on independent component analysis (ICA) was proposed. It is known that most of the ICA basis functions extracted from images are sparse and similar to localized and oriented receptive fields. In this paper, the L (p) norm is used to estimate sparseness of the ICA basis functions, and then, the sparser basis functions were selected for representing the edge information of an image. In the proposed method, a test image is first transformed by ICA basis functions, and then, the high-frequency information can be extracted with the components of the selected sparse basis functions. Furthermore, by applying a shrinkage algorithm to filter out the components of noise in the ICA domain, we can readily obtain the sparse components of the noise-free image, resulting in a kind of robust edge detection even for a noisy image with a very low SN ratio. The efficiency of the proposed method for edge detection is demonstrated by experiments with some medical images.

DOI： 10.1007/s11760-009-0140-5

J-GLOBAL

researchmap

Language：Japanese   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Successful liver surgery requires a clear understanding of the differences in liver shapes and vessel distribution in different individuals. Furthermore, in clinical medicine, there is a high demand for surgical assistance systems for individual patients. Therefore, we aim to segment the liver on the basis of the CT volume data, semi-automatically extract the vessels from the segmented livers and then visualize the 3D shape and the extracted vessel distribution using a virtual operation system. In addition, to improve the operability and accuracy of information recognition in the virtual operation system, prior knowledge and the clinical experiences of doctors are integrated into the visualization system for a practical virtual surgery. A 3D visualization of the liver, allows the user to easily recognize abnormal regions, which need to be removed, and to simply select this region using a 3D pointing device. Furthermore, 3D visualization, allows details in the structure of the human liver to be better understood and a more practical surgical simulation system can be implemented in our developed system.

CiNii Article

researchmap

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/mva/mva2011.html#conf/mva/HanCR11

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In this paper, we propose to use local feature set for image representation, which can represent variations in an object's appearance due to changing view point or camera pose. It was evidenced that usually only a part of the object are appeared in common when taking a photo of an object in different view points. With comparison of local features set extracted from different positions of images, an object can be recognized when common part is appeared in two images, which take photos of one object in different view points. In this paper, we use Canonical Correlation (also known as principle or canonical angles), which can be thought of as the angles between two d-dimensional subspace, as similarity measure of local feature sets. The proposed approach is evaluated in various view-based object datasets (Coil-100 and ETH80) for object and object category recognition. Experiments show that the performance advantages of our proposed approach can be achieved over existing techniques.

DOI： 10.1109/ICIP.2011.6116496

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Face Hallucination is, one of a learning-based super-resolution technique that can reconstruct a high-resolution image using only one low-resolution image. However, there are often some detailed high-frequency components of the reconstructed image that cannot be recovered using this method. In this study, we proposed a high-frequency compensated face hallucination method for enhancing reconstruction performance. The proposed method can be divided into three steps: 1) high-resolution image reconstruction using a conventional hallucination method; 2) residual (high-frequency components) image recovery by "training" a residual image pair; 3) compensation of the reconstructed high-resolution image obtained in step 1 with the reconstructed residual image. Experimental results show that the high-resolution images obtained using our proposed approach are much better than those obtained by conventional hallucination.

DOI： 10.1109/ICIP.2011.6115736

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

This paper addresses the problems of feature selection and feature fusion. For the feature selection, the color SIFT descriptors in the independent components are ordered for image classification. To select distinctive and compact independent components (IC) of the color SIFT descriptors, we propose two ordering approaches based on variation: (1) Local ordering approaches (the localization-based ICs ordering and the sparseness-based ICs ordering) and (2) Global selection approach (PCA-based ICs selection).We evaluate the performance of proposed methods on object and scene databases, and obtain the following two main results. First, the proposed methods are able to obtain acceptable classification results in comparison with original color SIFT descriptors. Second, the highest classification rate can be obtained by using the global selection method in the scene database, while the local ordering methods give the best performance for the object database. For the aspect of feature fusion, tensor-based ICA is utilized to consider the relationship between different features. This obtains compact and distinctive representation of images for effective image classification. © 2011 IEEE.

DOI： 10.1109/ICCP.2011.6047865

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

Subspace learning based face recognition methods have attracted considerable interest in recent years, including Principal Component Analysis (PCA), Independent Component Analysis (ICA), Linear Discriminant Analysis (LDA). and some extensions for 2D analysis. However, a disadvantage of all these approaches is that they perform subspace analysis directly on the reshaped vector or matrix of pixel-level intensity, which is usually unstable under illumination or pose variance. In this paper, we propose to represent a face image as a local descriptor tensor, which is a combination of the descriptor of local regions (K*K-pixel patch) in the image, and is more efficient than the popular Bag-Of-Feature (BOF) model for local descriptor combination. Furthermore, we propose to use a multilinear subspace learning algorithm (Supervised Neighborhood Embedding-SNF) for discriminant feature extraction from the local descriptor tensor of face images, which can preserve local sample structure in feature space. We validate our proposed algorithm on Benchmark database Yale and PIE, and experimental results show recognition rate with our method can be greatly improved compared conventional subspace analysis methods especially for small training sample number.

DOI： 10.1587/transinf.E94.D.158

researchmap

Language：English   Publishing type：Research paper (scientific journal)  

We describe an approach for the automatic modality classification in medical image retrieval task of the 2010 CLEF cross-language image retrieval campaign (ImageCLEF). This paper is focused on the process of feature extraction from medical images and fuses the different extracted visual features and textual feature for modality classification. To extract visual features from the images, we used histogram descriptor of edge, gray, or color intensity and block-based variation as global features and SIFT histogram as local feature. For textual feature of image representation, the binary histogram of some predefined vocabulary words from image captions is used. Then, we combine the different features using normalized kernel functions for SVM classification. Furthermore, for some easy misclassified modality pairs such as CT and MR or PET and NM modalities, a local classifier is used for distinguishing samples in the pair modality to improve performance. The proposed strategy is evaluated with the provided modality dataset by ImageCLEF 2010. © 2011 Xian-Hua Han and Yen-Wei Chen.

DOI： 10.1155/2011/241396

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

In this paper, we present a novel color independent components based SIFT descriptor (termed CIC-SIFT) for object/scene classification. We first learn an efficient color transformation matrix based on independent component analysis (ICA), which is adaptive to each category in a database. The ICA-based color transformation can enhance contrast between the objects and the background in an image. Then we compute CC-SIFT descriptors over all three transformed color independent components. Since the ICA-based color transformation can boost the objects and suppress the background, the proposed CIC-SIFT can extract more effective and discriminative local features for object/scene classification. The comparison is performed among seven SIFT descriptors, and the experimental classification results show that our proposed CC-SIFT is superior to other conventional SIFT descriptors.

DOI： 10.1587/transinf.E93.D.2577

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：IEICE-INST ELECTRONICS INFORMATION COMMUNICATIONS ENG  

In this paper, we present a novel color independent components based SIFT descriptor (termed CIC-SIFT) for object/scene classification. We first learn an efficient color transformation matrix based on independent component analysis (ICA), which is adaptive to each category in a database. The ICA-based color transformation can enhance contrast between the objects and the background in an image. Then we compute CC-SIFT descriptors over all three transformed color independent components. Since the ICA-based color transformation can boost the objects and suppress the background, the proposed CIC-SIFT can extract more effective and discriminative local features for object/scene classification. The comparison is performed among seven SIFT descriptors, and the experimental classification results show that our proposed CC-SIFT is superior to other conventional SIFT descriptors.

DOI： 10.1587/transinf.E93.D.2577

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：ELSEVIER SCIENCE BV  

Facial pose synthesis is applied to generate much required information for several applications, such as public security, facial cosmetology, etc. How to synthesize facial pose images from one image accurately without spatial information is still a challenging problem. In this paper we propose a tensor-based subspace learning method (TSL) for synthesizing human multi-pose facial images from a single two-dimensional image. In the proposed TSL method, two-dimensional multi-pose images in the database are previously organized into a tensor form and a tensor decomposition technique is applied to build projection subspaces. In synthesis processing, the input two-dimensional image is first projected into its corresponding projection subspace to get an identity vector and then the identity vector is used to generate other novel pose images. Our technique is applied on KAO-Ritsumeikan Multi-angle View, Illumination and Cosmetic Facial Database (MaVIC) and experiment results show the effectiveness of our proposed method for facial pose synthesis. (C) 2010 Elsevier B.V. All rights reserved.

DOI： 10.1016/j.neucom.2010.04.013

J-GLOBAL

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Super-resolution (SR) enhancement from multi-frame low-resolution (LR) images (multi-frame superresolution) has been a well-studied topic in the literature. Image registration is the most important part for multi-frame super-resolution, and accurate alignment of LR images would contribute a critical role for the final success of SR image reconstruction. In this paper, we propose to combine the Principle Component Analysis (PCA) based registration method, which can perform object alignment in real-time and without constraints on the three registration parameters (i.e., translation, rotation, and scaling), and gradient registration method, which can perform precise registration with minor image movement. Experimental results show that the reconstruction SR images by our proposed method have much higher quality than those by the state of art algorithms.

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

In this paper, we develop a new method for prediction O-linked glycosylation site and pattern analysis in protein, which combines independent component analysis (ICA) with a multi-layer neural network (NN). ICA is first used to. construct main basis (subspace) of the protein sequence for features extraction. The projections of protein sequence on the subspace with low dimension are used as input data instead of the higher-dimensional protein sequences. Neural network is built to predict whether a particular site of serine or threonine is glycosylated. Compared with other subspace method, our proposed new method can improve the prediction accuracy.

researchmap

Publisher：National Institute of Standards and Technology (NIST)  

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/trecvid/trecvid2010.html#conf/trecvid/AiAKYHC10

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

In this paper, we propose a multilinear (N-Dimensional) Tensor Supervised Neighborhood Embedding (called ND-TSNE) for discriminant feature representation, which is used for view-based object recognition. ND-TSNE use a general Nth order tensor discriminant and neighborhood-embedding analysis approach for object representation. The benefits of ND-TSNE include: (1) a natural way of representing data without losing structure information, i.e., the information about the relative positions of pixels or regions; (2) a reduction in the small sample size problem which occurs in conventional supervised learning because the number of training samples is much less than the dimensionality of the feature space; (3) a neighborhood structure preserving in tensor feature space for object recognition and a good convergence property in training procedure. With Tensor-subspace features, the random forests as a multi-way classifier is used for object recognition, which is much easier for training and testing compared with multi-way SVM. We demonstrate the performance advantages of our proposed approach over existing techniques using experiments on the COIL-100 and the ETH-80 datasets.

DOI： 10.1007/978-3-642-15702-8_22

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

In this paper, we present a novel semi-supervised and interactive concept learning algorithm for scene recognition by local semantic description. Our work is motivated by the continuing effort in content-based image retrieval to extract and to model the semantic content of images. The basic idea of the semantic modeling is to classify local image regions into semantic concept classes such as water, sunset, or sky [1]. However, labeling concept sampling manually for training semantic model is fairly expensive, and the labeling results is, to some extent, subjective to the operators. In this paper, by using the proposed semi-supervised and interactive learning algorithm, training samples and new concepts can be obtained accurately and efficiently. Through extensive experiments, we demonstrate that the image concept representation is well suited for modeling the semantic content of heterogenous scene categories, and thus for recognition and retrieval. Furthermore,higher recognition accuracy can be achieved by updating new training samples and concepts, which are obtained by the novel proposed algorithm. © 2010 IEEE.

DOI： 10.1109/ICPR.2010.746

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

Image category recognition is important to access visual information on the level of objects and scene types. This paper presents a new algorithm for the automatic recognition of object and scene classes. Compact and yet discriminative visual-words and low-level-features object class subspaces are automatically learned from a set of training images by a Supervised Nonlinear Neighborhood Embedding (SNNE) algorithm, which can learn an adaptive nonlinear subspace by preserving the neighborhood structure of the visual feature space. The main contribution of this paper is two fold: i) an optimally compact and discriminative feature subspace is learned by the proposed SNNE algorithm for different feature space (visual-word and low-level features). ii) An effective merge of different feature subspace can be implemented simply. High classification accuracy is demonstrated on different database including the scene database (SIMPLicity) and object recognition database (Caltech). We confirm that the proposed strategy is much better than state-of-the-art methods for different databases. © 2010 IEEE.

DOI： 10.1109/ICPR.2010.744

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

This paper proposes an adaptive color independent components based SIFT descriptor (termed CIC-SIFT) for image classification. Our motivation is to seek an adaptive and efficient color space for color SIFT feature extraction. Our work has two key contributions. First, based on independent component analysis (ICA), an adaptive and efficient color space is proposed for color image representation. Second, in this ICA-based color space, a discriminative CIC-SIFT descriptor is calculated for image classification. The experiment results indicate that (1) contrast between objects and background can be enhanced on the ICA-based color space and (2) the CIC-SIFT descriptor outperforms other conventional color SIFT descriptors on image classification. © 2010 IEEE.

DOI： 10.1109/ICPR.2010.596

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Image category recognition is important to access visual information on the level of objects and scene types. This paper combines different feature representations of images and learn a compact subspace of different features for the automatic recognition of object and scene classes. Compact visual-words and low-level-features object class subspaces are automatically learned from a set of training images by a Regularized Linear Discriminant analysis (RLDA) algorithm, and the extracted RLDA-domain features are used for Support Vector Machine (SVM) classifier. The main contribution of this paper is two folds: i) Different features (bag-of-features and low-level features) is fused for image representation. ii) The compact feature subspaces (low-dimension features) of different features are learned for rendering to SVM classifier, which is computationally efficient for image category. High classification accuracy is demonstrated on object recognition database (Caltech). We confirm that the proposed strategy cam improve accuracy rate compared with state-of-the-art methods for object recognition databases.

DOI： 10.1109/ICIP.2010.5653931

researchmap

Publisher：CEUR-WS.org  

researchmap

Other Link： http://dblp.uni-trier.de/db/conf/clef/clef2010w.html#conf/clef/Chen10

Language：English   Publishing type：Research paper (scientific journal)   Publisher：ICIC INTERNATIONAL  

Since the MR signals can be considered as a combination of the signals from each brain matters, it has been shown that independent component analysis (ICA) con be used for contrast enhancement of MR images. However, ICA is a linear method in nature, and it is inadequate to well-describe nonlinear variations of the real MR images. In this paper, we propose a new method for contrast enhancement of MR brain images using a canonical correlation based kernel independent component analysis (KICA). Experimental results on both phantom MR datasets and real clinical MR datasets show that the contrast of MR images can be significantly enhanced by KICA.

researchmap

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Facial pose synthesis has many useful applications in practice. How to synthesize facial pose images robustly and simply is still a challenging problem. In this paper we proposed a tensor-based subspace learning method (TSL) that makes possible the synthesis of human multi-pose facial images from a single 2D image. We organize 2D multi-pose images in a tensor form and apply tensor decomposition to build a projection subspace. An input 2D image is projected into the projection subspace to get a corresponding identity vector. The identity vector is used to generate the novel pose images. The experiments are performed on MaVIC (KAO-Ritsumeikan Multi-angle View, Illumination and Cosmetic Facial Database) and preliminary experimental results show the effectiveness of our proposed method. ©2009 IEEE.

DOI： 10.1109/ICCVW.2009.5457697

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

We proposed a new feature extraction method based on supervised locality preserving projections (SLPP) for region segmentation and categorization in high-resolution satellite images. Compared with other subspace methods such as PCA and ICA, SLPP can preserve local geometric structure of data and enhance within-class local information. The generalization of the proposed SLPP based method is discussed in this paper.

DOI： 10.1007/978-3-642-03265-3_24

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

Image category recognition is important to access visual information on the level of objects and scene types. This paper combines different feature representations of images and learn a compact subspace of different features for the automatic recognition of object and scene classes. Compact visual-words and low-level-features object class subspaces are automatically learned from a set of training images by a Principle Component Analysis (PCA) algorithm, and the extracted PCA-domain features are used for Support Vector Machine (SVM) classifier. The main contribution of this paper is two fold: i) Different features (bag-of-features and low-level features)is fused for image representation. ii) The compact feature subspaces (low-dimension features) of different features are learned for rendering to SVM classifier, which is computationally efficient for image category. High classification accuracy is demonstrated on object recognition database (Caltech). We confirm that the proposed strategy is comparable with state-of-the-art methods for object recognition databases. © 2009 IEEE.

DOI： 10.1109/IIH-MSP.2009.31

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

Glycosylation is one of the common post-translation modification of protein in eukaryotic cells. Conventional neural network methods have been applied to predict glycosylation sites in protein sequence and the prediction accuracy is dependent on the dimension of feature vector (length of protein sequence). Though the prediction accuracy can be improved by increasing the length of protein sequence, it is time-consuming. In this paper, we propose a novel approach which combines PCA with a multilayer neural network for efficient, and accurate prediction of O-glycosylation sites in protein. PCA is first used to extract main basis (subspace) of the protein sequence. The lower-dimensional projections on the subspace are used as features instead of the higher-dimensional protein sequences. Compared with conventional method (without PCA), our proposed method can significantly reduce the larger computation cost, while can keep the higher prediction accuracy. © 2009 IEEE.

DOI： 10.1109/IIH-MSP.2009.182

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

Image category recognition is important to access visual information on the level of objects and scene types. This paper presents an automatic recognition system of scene and object with PCA-SICEF feature for digital color images. SICEF (Scale-Invariant Color and edge Feature) is an extension of the conventional local SIFT (Scale-Invariant Feature transform) feature, which only include edge invariance of local image region but not any color information. So the SIFT feature is not enough for distinguish image categorization especially for scene types, where the color information plays an important role for recognition. Therefore, we improve SIFT by including color feature for local image region, and name it as SICEF feature. However, the Dimension of the extracted SICEF feature is so high that we use PCA (Principle Component Analysis) to reduce the dimension, and then, use the PCA-domain SICEF (PCA-SICEF) for image classification. Experimental results show that it is much more efficient by our proposed PCA-SICEF feature than conventional SIFT feature. © 2009 IEEE.

DOI： 10.1109/ICNC.2009.655

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE Computer Society  

Recently, with the development of visual communication and image processing, there is a high demand for High-Resolution images such as video surveillance, medical imaging, and so on. Therefore, the Super-Resolution technology that produces a highresolution image from a set of shified, blurred, and decimated versions is actively researched. However, most previously published techniques perform well only for small magnifications but get worse either in computational complexity or ringing artifacts for large magnifications. In this paper, we propose a Hierarchical algorithm for high-magnification SuperResolution image reconstruction. The proposed algorithm magnifies the low-resolution image in multiple steps. Experiment results show that the new approach is more efficient and can provide much better reconstruction quality in comparison with the reconstruction result in one step. © 2009 IEEE.

DOI： 10.1109/ICNC.2009.658

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：ACM  

This paper develops a supervised nonlinear subspace of bag-of-features for category classification. Bag-of-features represents an image as an orderless distribution of features, which selects the visual words by clustering and uses the similarity with each visual word as the features for classification. In this paper, we propose to model the ensemble of visual words with a supervised nonlinear neighborhood embedding method to a more discriminative space for category classification. The supervised nonlinear neighborhood embedding(SNNE) is used to model visual words and extract the discrimitive information specialized for each category. The projection length in subspace is used as features for classification. The SNNE subspace method can model the nonlinear variations induced by various kinds of visual words and extract more discriminative feature for object recognition. The proposed method is evaluated using the Cal-tech and GRAZ01 database. We confirm that the proposed method is comparable with state-of-the-art methods without absolute position information. Copyright 2009 ACM.

DOI： 10.1145/1734605.1734615

Scopus

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

This paper develops a supervised nonlinear subspace of bag-of-features for category classification. Bag-of-features represents an image as an orderless distribution of features, which selects the visual words by clustering and uses the similarity with each visual word as the features for classification. In this paper, we propose to model the ensemble of visual words with a supervised nonlinear neighborhood embedding method to a more discriminative space for category classification. The supervised nonlinear neighborhood embedding(SNNE) is used to model visual words and extract the discrimitive information specialized for each category. The projection length in subspace is used as features for classification. The SNNE subspace method can model the nonlinear variations induced by various kinds of visual words and extract more discriminative feature for object recognition. The proposed method is evaluated using the Cal-tech and GRAZ01 database. We confirm that the proposed method is comparable with state-of-the-art methods without absolute position information. Copyright 2009 ACM.

DOI： 10.1145/1734605.1734615

Scopus

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：SCIENCE PRESS  

We propose a robust edge detection method based on ICA-domain shrinkage (independent component analysis). It is known that most basis functions extracted from natural images by ICA are sparse and similar to localized and oriented receptive fields, and in the proposed edge detection method, a target image is first transformed by ICA basis functions and then the edges are detected or reconstructed with sparse components. Furthermore, by applying a shrinkage algorithm to filter out the components of noise in ICA-domain, we can readily obtain the sparse components of the original image, resulting in a kind of robust edge detection even for a noisy image with a very low SN ratio. The efficiency of the proposed method is demonstrated by experiments with some natural images.

DOI： 10.1007/s11432-008-0114-1

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：ICIC INT  

In this paper, a spatio-chromatic ICA based noise reduction in color images is presented. We use RGB color images with a regular arrangement RGB channels into spatio-chromatic space, and ICA performs decorrelation of these signals. We show that spatio-chromatic components of images contain spatial information and color information, and that ICA spatio-chromatic analysis is able to help the reconstruction of images. We propose an efficient noise reduction method for color images based on the observation. In the proposed method, the image is first transformed to spatio-chromatic ICA domain and then the noise components are removed by soft thresholding (Shrinkage). Experimental results show that the denoised images based spatio-chromatic ICA domain shrinkage are dramatically improved in comparison to that with conventional filtering.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Subspace learning. techniques are widespread in pattern recognition research. They include PCA, ICA, LPP, etc. These techniques are generally linear and unsupervised. The problem of image indexing is very complicated and the processed images are usually lie on non-linear image subspaces. In this paper, we propose. a supervised nonlinear neighborhood embedding: algorithm which learns an adaptive nonlinear subspace by preserving: the neighborhood structure of the image color.,pace. In the proposed algorithm, we combine the idea of nonlinear kernel mapping and preserving the neighborhood structure of the sampled, so it can not only gain a perfect approximation of the nonlinear image manifold, but also enhance within-class neighborhood information. Experimental results show that the proposed method outperform other linear or unsupervised subspace learning, methods.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPRINGER-VERLAG BERLIN  

We proposed a new method based on Supervised locality preserving projections (SLPP) for classification of high resolution satellite images. Compared with other subspace methods Such as PCA and ICA, SLPP can preserve local geometric structure of data and enhance within-class local information. The proposed method has been successfully applied to IKONOS images and experimental results show that the proposed SLPP based method Outperform ICA-based method. The proposed method can be practically incorporated into a GIS system.

DOI： 10.1007/978-3-540-85565-1-19

DOI： 10.1007/978-3-540-85565-1_19

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

It is important to enhance and detect nodules in CT images in order to identify the lung cancer at early stage. Nodules which are included in medical image generally have multiple size and scale, and have blob-like structure. Recently, 3D multiscale filter approach is proposed for lung nodules detection. However, the 3D method take too much computing time to be applicable in real diagnose systems. In this paper; we propose a two step method for lung nodules detection in CT images. Firstly, we use 2D multiscale filter to detect the candidates of lung nodules on the slice images, and then, reduce most of false positive nodules using logical AND operator of continuous CT slices. The effectiveness of the purposed methods has been demonstrated by Experiments.

DOI： 10.1109/IIH-MSP.2008.303

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

This paper presents an auto-annotation system with simple pre-processed segmentation for digital color image. Recently, annotation techniques become one popular method for image retrieval system in image database management[1], image recognition system[2][3] and so on. In the paper, we propose a two-step approach for image annotation. Firstly, the color image is needed to be segmented into two parts: the main object is assumed as the foreground part, and the other will be the background part, and then, only the features of the foreground part (main object) are used for annotation of global images. Here, it is assumed that only a single main object is included in the color image, so annotation problem can be considered as a kind of classification problem, and all the images in database can be automatically categorized by neural network. After image annotating, user can easily retrieve the similar set of images with the same conception that user needs. Finally, we compare the performance of the proposed systems with different processing methods. Experimental results show efficiency of the proposed annotation system.

DOI： 10.1109/IIH-MSP.2008.338

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

At present there are many methods that could deal well with frontal view face recognition when there is sufficient number of representative training samples. Thereinto, sub-space learning method such as Principal component analysis(PCA), Independent component analysis(ICA), Linear Discriminant Analysis(LDA) are a very hot research topic in this field. However, in some face recognition system, the needed recognition faces are different with different users and are often changed according users' requirement. So in this paper, we proposed to make use of some known face database, in which subjects will be different with the test subjects, for training(extract subspace) with a modified ICA method. The proposed modified ICA method can save much of computer time and memory, and at the same time, can obtain acceptable experimental results on a part of FERET face database. In addition, we validate that the accuracy rate with simple rotation images of logged faces as known data can be improved when the logged face is only one sample per subject.

DOI： 10.1109/IIH-MSP.2008.222

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Subspace learning techniques are widespread in pattern recognition research. They include PCA, ICA, LPP, etc. These techniques are generally linear and unsupervised. The problem of image indexing is very complicated and the processed images are usually lie on non-linear image subspaces. In this paper, we propose a supervised nonlinear neighborhood embedding algorithm which learns an adaptive nonlinear subspace by preserving the neighborhood structure of the image color space. In the proposed algorithm, we combine the idea of nonlinear kernel mapping and preserving the neighborhood structure of the samples, so it can not only gain a perfect approximation of the nonlinear image manifold, but also enhance within-class neighborhood information. Experimental results show that the proposed method outperform other linear or unsupervised subspace learning methods.

DOI： 10.1109/ICIP.2008.4711913

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE-INT SOC OPTICAL ENGINEERING  

CT (computed tomography) imaging is a technology which uses X-ray beams (radiation) and computers to form detailed, cross-sectional images of an area of anatomy. However, the random scattered X-ray in CT imaging system will reduce radiographic contrast greatly in CT images. In this paper, a four-step method is proposed for decoding CT images: first, the EGSnrc Monte Carlo simulation system is used to simulate CT imaging and simulated data will be validated by real experimental data in the same experimental conditions; second, scattered X-ray image simulated by EGSnrc will be transformed into ICA-domain (independent component analysis-domain) to obtain the main magnitude of scattering data; third, a noise-reduction algorithm based on ICA-domain shrinkage is applied to smooth the CT image; fourth, the conventional linear deconvolution follows. The simulation results show that the reconstructed image is dramatically improved in comparison to that without the noise-removing filters, and the proposed method is also applied to real experimental X-ray imaging.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Cone beam CT has a capability for the 3-dimentional imaging of large volumes with isotropic resolution. But the main limitation of cone beam CT is a larger amount of x-ray scatter. The x-ray scatter may enhance the noise in the reconstructed images. In this paper, we propose a new method based on independent component analysis (ICA) for removing the x-ray scatter in the observed x-ray images. The observed x-ray images can be considered as a linear mixture of scattered x-ray and primary x-ray. We show that the scattered x-ray and the primary x-ray can be blindly separated from two observed x-ray images by the use of ICA. The results indicate that the proposed method is effective for removing x-ray scatter from the x-ray images.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE  

Projection data in Positron Emission Tomography (PET) are acquired as a number of photon counts from different observation angles. Positron decay is a random phenomenon that causes undesirably high variations in measured sinogram appearing as quantum noise. The ruduction of quantum noise or Poisson noise in medical images is an important issue. In this paper, we propose a new ICA-based filter for reduction of noise in sinogram domain. In the proposed filter, the sinogram (projection data) is firstly transformed to ICA domain, and then, the components of scattered projection are removed by a soft thresholding (Shrinkage). In this study, the choice of ICA basis function trained from different database is considered. The denoised results with different ICA basis fuinctions and conventional denoising method(wavelet shrinkage and Gaussian filter) are given for comparison, and then, we also show the reconstructed images of ICA-based denoised sinogram images using Filtered-Back-Projection(FBP) algorithm. Experimental results show that the reconstructed images of ICA-based denoised images are much clearer and have much better contrast than those without pre-processing filters.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE-INT SOC OPTICAL ENGINEERING  

Interventional Radiology (IVR) is an important technique to visualize and diagnosis the vascular disease. In real medical application, a weak x-ray radiation source is used for imaging in order to reduce the radiation dose, resulting in a low contrast noisy image. It is important to develop a method to smooth out the noise while enhance the vascular structure. In this paper, we propose to combine an ICA Shrinkage filter with a multiscale filter for enhancement of IVR images. The ICA shrinkage filter is used for noise reduction and the multiscale filter is used for enhancement of vascular structure. Experimental results show that the quality of the image can be dramatically improved without any blurring in edge by the proposed method. Simultaneous noise reduction and vessel enhancement have been achieved.

DOI： 10.1117/12.752036

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：SPIE-INT SOC OPTICAL ENGINEERING  

Penumbral imaging is a kind of technique which uses the facts that spatial information can be recovered from the shadow or penumbra that an unknown source casts through a simple large circular aperture. The technique is based on a linear deconvolution. As we known that the information of the penumbral image is only contained in the penumbra (the edges of the image), so according to that principle, we proposed a two-step method for decoding penumbral images in this paper. First, an edgy-emphasizing algorithm using a band filter is applied to extract the penumbras (the image edges) in noisy penumbral images; then, followed by conventional linear deconvolution of only the penumbral edges. The simulation results show that the reconstructed image is dramatically improved in comparison to that with the conventional noise-removing filters, and the proposed method is also applied to real experimental x-ray imaging.

DOI： 10.1117/12.740386

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

The reduction of noise in medical images is an important issue. In this paper we propose a new ICA-based filter for reduction of noise in reconstruction domain. In the proposed filter the reconstructed 3D PET images(X- Y plane-slice domain or X-Z plane) are firstly transformed to ICA domain, and then, the components of noise information are removed by a soft thresholding (Shrinkage). In this study, the choice of ICA basis function trained from noisy reconstructed images in different plane is considered. The denoised results with different ICA basis functions and conventional denoising method(wavelet shrinkage and Gaussian filter) are given for comparison. Experimental results show that the reconstructed images of ICA-based denoised images are much clearer and have much better contrast than those with wavelet-domain filters.

DOI： 10.1109/IIH-MSP.2007.87

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)   Publisher：IEEE COMPUTER SOC  

This paper proposes a new method based on independent component analysis (ICA) for Poisson noise reduction. In the proposed method, the image is first transformed to ICA domain and then the noise components are removed by a soft thresholding (Shrinkage). The proposed method, which is used as a preprocessing of the reconstruction, has been successfully applied to penumbral imaging. Both simulation results and experimental results show that the reconstructed image is dramatically improved in comparison to that without the noise-removing filters.

DOI： 10.1109/FSKD.2007.180

researchmap

researchmap

Language：English   Publishing type：Research paper (scientific journal)   Publisher：ICIC INTERNATIONAL  

We propose a robust edge extraction algorithm based on Principal component analysis (PCA) for Poisson noise reduction. In the proposed edge detection, the image is firstly transformed to PCA subspace with sparse PCA basis functions and then the noisy components are removed by a soft threshold (Shrinkage). The proposed edge extraction method, which is used as a preprocessing step of the reconstruction, has been successfully applied to penumbral imaging. Both simulation results and real experimental results show that the reconstructed images are dramatically improved in comparison to that with the conventional noise-removing filter.

researchmap

Language：English   Publishing type：Research paper (international conference proceedings)  

Radiological imaging such as x-ray CT is one of the most important tools for medical diagnostics. Since the radiological images are always with some quantum noise and the reduction of quantum noise or Poisson noise in medical images is an important issue. In this paper, we propose a new filtering based on independent component analysis (ICA) for reduction of noise. In the proposed filtering, the image (projection) is first transformed to ICA domain and then the components of scattered x-ray are removed by a soft thresholding (Shrinkage). The proposed method has been demonstrated by using both standard images and Monte Carlo simulations. Experimental results show that the quality of the image can be dramatically improved without any blurring in edge by the proposed filter.

DOI： 10.1117/12.652043