Over the last few decades, Gaussian Mixture Model (GMM) has attracted considerable interest in data mining and pattern recognition. GMM can be used to cluster a bunch of data through estimating the parameters of multiple Gaussian components using Expectation-Maximization (EM). Recently, Locally Consistent GMM (LCGMM) has been proposed to improve the clustering performance of GMM by exploiting the local manifold structure modeled by a p nearest neighbor graph. In practice, various prior knowledge may be available which can be used to guide the clustering process and improve the performance. In this paper, we introduce a semi-supervised method, called Semi-supervised LCGMM (Semi-LCGMM), where prior knowledge is provided in the form of class labels of partial data. Semi-LCGMM incorporates prior knowledge into the maximum likelihood function of LCGMM and is solved by EM. It is worth noting that in our algorithm each class has multiple Gaussian components while in the unsupervised settings each class only has one Gaussian component. Experimental results on several datasets demonstrate the effectiveness of our algorithm.
{"title":"Manifold Regularized Gaussian Mixture Model for Semi-supervised Clustering","authors":"Haitao Gan, N. Sang, Rui Huang, X. Chen","doi":"10.1109/ACPR.2013.126","DOIUrl":"https://doi.org/10.1109/ACPR.2013.126","url":null,"abstract":"Over the last few decades, Gaussian Mixture Model (GMM) has attracted considerable interest in data mining and pattern recognition. GMM can be used to cluster a bunch of data through estimating the parameters of multiple Gaussian components using Expectation-Maximization (EM). Recently, Locally Consistent GMM (LCGMM) has been proposed to improve the clustering performance of GMM by exploiting the local manifold structure modeled by a p nearest neighbor graph. In practice, various prior knowledge may be available which can be used to guide the clustering process and improve the performance. In this paper, we introduce a semi-supervised method, called Semi-supervised LCGMM (Semi-LCGMM), where prior knowledge is provided in the form of class labels of partial data. Semi-LCGMM incorporates prior knowledge into the maximum likelihood function of LCGMM and is solved by EM. It is worth noting that in our algorithm each class has multiple Gaussian components while in the unsupervised settings each class only has one Gaussian component. Experimental results on several datasets demonstrate the effectiveness of our algorithm.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125865126","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Optical Character Recognition (OCR) in video stream of flipping pages is a challenging task because flipping at random speed cause difficulties to identify frames that contain the open page image (OPI) for better readability. Also, low resolution, blurring effect shadows add significant noise in selection of proper frames for OCR. In this work, we focus on the problem of identifying the set of optimal representative frames for the OPI from a video stream of flipping pages and then perform OCR without using any explicit hardware. To the best of our knowledge this is the first work in this area. We present an algorithm that exploits cues from edge information of flipping pages. These cues, extracted from the region of interest (ROI) of the frame, determine the flipping or open state of a page. Then a SVM classifier is trained with the edge cue information for this determination. For each OPI we obtain a set of frames. Next we choose the central frame from that set of frames as the representative frame of the corresponding OPI and perform OCR. Experiments are performed on video documents recorded using a standard resolution camera to validate the frame selection algorithm and we have obtained 88% accuracy. Also, we have obtained character recognition accuracy of 82% and word recognition accuracy of 77% from such book flipping OCR.
{"title":"OCR from Video Stream of Book Flipping","authors":"Dibyayan Chakraborty, P. Roy, J. Álvarez, U. Pal","doi":"10.1109/ACPR.2013.24","DOIUrl":"https://doi.org/10.1109/ACPR.2013.24","url":null,"abstract":"Optical Character Recognition (OCR) in video stream of flipping pages is a challenging task because flipping at random speed cause difficulties to identify frames that contain the open page image (OPI) for better readability. Also, low resolution, blurring effect shadows add significant noise in selection of proper frames for OCR. In this work, we focus on the problem of identifying the set of optimal representative frames for the OPI from a video stream of flipping pages and then perform OCR without using any explicit hardware. To the best of our knowledge this is the first work in this area. We present an algorithm that exploits cues from edge information of flipping pages. These cues, extracted from the region of interest (ROI) of the frame, determine the flipping or open state of a page. Then a SVM classifier is trained with the edge cue information for this determination. For each OPI we obtain a set of frames. Next we choose the central frame from that set of frames as the representative frame of the corresponding OPI and perform OCR. Experiments are performed on video documents recorded using a standard resolution camera to validate the frame selection algorithm and we have obtained 88% accuracy. Also, we have obtained character recognition accuracy of 82% and word recognition accuracy of 77% from such book flipping OCR.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124137927","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this paper, a graph-based off-line handwritten signature verification system is proposed. The system can automatically identify some global and local features which exist within different signatures of the same person. Based on these features it is possible to verify whether a signature is a forgery or not. The structural description in the form of hierarchical attributed random graph set is transformed into matrix-vector structures. These structures can be directly used as matching pattern when examined signature is analyzed. The proposed approach can be applied to off-line signature verification systems especially for kanji-like or ideogram-based structurally complex signatures.
{"title":"Matrix-Based Hierarchical Graph Matching in Off-Line Handwritten Signatures Recognition","authors":"M. Piekarczyk, M. Ogiela","doi":"10.1109/ACPR.2013.164","DOIUrl":"https://doi.org/10.1109/ACPR.2013.164","url":null,"abstract":"In this paper, a graph-based off-line handwritten signature verification system is proposed. The system can automatically identify some global and local features which exist within different signatures of the same person. Based on these features it is possible to verify whether a signature is a forgery or not. The structural description in the form of hierarchical attributed random graph set is transformed into matrix-vector structures. These structures can be directly used as matching pattern when examined signature is analyzed. The proposed approach can be applied to off-line signature verification systems especially for kanji-like or ideogram-based structurally complex signatures.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"96 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131423890","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this study, we show the mobile photographing robot which moves around entertainment facilities, and automatically takes a picture of people who desire a commemorative picture. Our robot approaches target human by detecting his/her face from the image captured from monocular camera attached on the robot. In our method, the robot behaviors are controlled by using the face detection results. In order to validate the usefulness of the proposed method, the performance of the mobile photographing robot is evaluated. From the experimental results, it was shown that the robot could approach a human and take a picture automatically without operator's intervention from human approaching to photographing.
{"title":"Mobile Robot Photographer","authors":"Satoru Suzuki, Y. Mitsukura","doi":"10.1109/ACPR.2013.192","DOIUrl":"https://doi.org/10.1109/ACPR.2013.192","url":null,"abstract":"In this study, we show the mobile photographing robot which moves around entertainment facilities, and automatically takes a picture of people who desire a commemorative picture. Our robot approaches target human by detecting his/her face from the image captured from monocular camera attached on the robot. In our method, the robot behaviors are controlled by using the face detection results. In order to validate the usefulness of the proposed method, the performance of the mobile photographing robot is evaluated. From the experimental results, it was shown that the robot could approach a human and take a picture automatically without operator's intervention from human approaching to photographing.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126493203","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Person re-identification is becoming a hot research topic due to its academic importance and attractive applications in visual surveillance. This paper focuses on solving the relatively harder and more importance multiple-shot re-identification problem. Following the idea of treating it as a set-based classification problem, we propose a new model called Locality-constrained Collaborative Sparse Approximation (LCSA) which is made to be as efficient, effective and robust as possible. It improves the very recently proposed Collaborative Sparse Approximation (CSA) model by introducing two types of locality constraints to enhance the quality of the data for collaborative approximation. Extensive experiments demonstrate that LCSA is not only much better than CSA in terms of effectiveness and robustness, but also superior to other related methods.
{"title":"Locality-Constrained Collaborative Sparse Approximation for Multiple-Shot Person Re-identification","authors":"Yang Wu, M. Mukunoki, M. Minoh","doi":"10.1109/ACPR.2013.14","DOIUrl":"https://doi.org/10.1109/ACPR.2013.14","url":null,"abstract":"Person re-identification is becoming a hot research topic due to its academic importance and attractive applications in visual surveillance. This paper focuses on solving the relatively harder and more importance multiple-shot re-identification problem. Following the idea of treating it as a set-based classification problem, we propose a new model called Locality-constrained Collaborative Sparse Approximation (LCSA) which is made to be as efficient, effective and robust as possible. It improves the very recently proposed Collaborative Sparse Approximation (CSA) model by introducing two types of locality constraints to enhance the quality of the data for collaborative approximation. Extensive experiments demonstrate that LCSA is not only much better than CSA in terms of effectiveness and robustness, but also superior to other related methods.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125265722","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jing Li, Zhichao Lu, Gang Zeng, Rui Gan, Long Wang, H. Zha
Depth map super resolution from multi-view depth or color images has long been explored. Multi-view stereo methods produce fine details at texture areas, and depth recordings would compensate when stereo doesn't work, e.g. at non-texture regions. However, resolution of depth maps from depth sensors are rather low. Our objective is to produce a high-res depth map by fusing different sensors from multiple views. In this paper we present a learning-based method, and infer a high-res depth map from our synthetic database by minimizing the proposed energy. As depth alone is not sufficient to describe geometry of the scene, we use additional features like normal and curvature, which are able to capture high-frequency details of the surface. Our optimization framework explores multi-view depth and color consistency, normal and curvature similarity between low-res input and the database and smoothness constraints on pixel-wise depth-color coherence as well as on patch borders. Experimental results on both synthetic and real data show that our method outperforms state-of-the-art.
{"title":"A Joint Learning-Based Method for Multi-view Depth Map Super Resolution","authors":"Jing Li, Zhichao Lu, Gang Zeng, Rui Gan, Long Wang, H. Zha","doi":"10.1109/ACPR.2013.89","DOIUrl":"https://doi.org/10.1109/ACPR.2013.89","url":null,"abstract":"Depth map super resolution from multi-view depth or color images has long been explored. Multi-view stereo methods produce fine details at texture areas, and depth recordings would compensate when stereo doesn't work, e.g. at non-texture regions. However, resolution of depth maps from depth sensors are rather low. Our objective is to produce a high-res depth map by fusing different sensors from multiple views. In this paper we present a learning-based method, and infer a high-res depth map from our synthetic database by minimizing the proposed energy. As depth alone is not sufficient to describe geometry of the scene, we use additional features like normal and curvature, which are able to capture high-frequency details of the surface. Our optimization framework explores multi-view depth and color consistency, normal and curvature similarity between low-res input and the database and smoothness constraints on pixel-wise depth-color coherence as well as on patch borders. Experimental results on both synthetic and real data show that our method outperforms state-of-the-art.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"146 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116616173","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Jiangyong Duan, Gaofeng Meng, Shiming Xiang, Chunhong Pan
This paper presents a new method for multifocus image fusion. We formulate the problem as an optimization framework with three terms to model common visual artifacts. A reconstruction error term is used to remove the boundary seam artifacts, and an out-of-focus energy term is used to remove the ringing artifacts. Together with an additional smoothness term, these three terms define the objective function of our framework. The objective function is then minimized by an efficient greedy iteration algorithm. Our method produces high quality fusion results with few visual artifacts. Comparative results demonstrate the efficiency of our method.
{"title":"Multifocus Image Fusion via Region Reconstruction","authors":"Jiangyong Duan, Gaofeng Meng, Shiming Xiang, Chunhong Pan","doi":"10.1109/ACPR.2013.92","DOIUrl":"https://doi.org/10.1109/ACPR.2013.92","url":null,"abstract":"This paper presents a new method for multifocus image fusion. We formulate the problem as an optimization framework with three terms to model common visual artifacts. A reconstruction error term is used to remove the boundary seam artifacts, and an out-of-focus energy term is used to remove the ringing artifacts. Together with an additional smoothness term, these three terms define the objective function of our framework. The objective function is then minimized by an efficient greedy iteration algorithm. Our method produces high quality fusion results with few visual artifacts. Comparative results demonstrate the efficiency of our method.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":" 9","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120830139","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hand gesture segmentation is the task of interpreting and spotting meaningful hand gestures from continuous hand gesture sequences with non-sign transitional hand movements. In real world scenarios, challenges from the unconstrained environments can largely affect the performance of gesture segmentation. In this paper, we propose a gesture spotting scheme which can detect and monitor all eligible hand candidates in the scene, and evaluate their movement trajectories with a novel method called Partition Matrix based on Hidden Conditional Random Fields. Our experimental results demonstrate that the proposed method can spot meaningful hand gestures from continuous gesture stream with 2-4 people randomly moving around in an uncontrolled background.
{"title":"Hand Gesture Segmentation in Uncontrolled Environments with Partition Matrix and a Spotting Scheme Based on Hidden Conditional Random Fields","authors":"Yi Yao, Chang-Tsun Li","doi":"10.1109/ACPR.2013.153","DOIUrl":"https://doi.org/10.1109/ACPR.2013.153","url":null,"abstract":"Hand gesture segmentation is the task of interpreting and spotting meaningful hand gestures from continuous hand gesture sequences with non-sign transitional hand movements. In real world scenarios, challenges from the unconstrained environments can largely affect the performance of gesture segmentation. In this paper, we propose a gesture spotting scheme which can detect and monitor all eligible hand candidates in the scene, and evaluate their movement trajectories with a novel method called Partition Matrix based on Hidden Conditional Random Fields. Our experimental results demonstrate that the proposed method can spot meaningful hand gestures from continuous gesture stream with 2-4 people randomly moving around in an uncontrolled background.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116495093","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Akinobu Maejima, A. Mizokawa, Daiki Kuwahara, S. Morishima
We propose a facial aging simulator which can synthesize a photo realistic human aged-face image for criminal investigation. Our aging simulator is based on the patch-based facial texture reconstruction with a wrinkle aging pattern model. The advantage of our method is to synthesize an aged-face image with detailed skin texture such as spots and somberness of facial skin, as well as age-related facial wrinkles without blurs that are derived from lack of accurate pixel-wise alignments as in the linear combination model, while maintaining the identity of the original face.
{"title":"Facial Aging Simulator Based on Patch-Based Facial Texture Reconstruction","authors":"Akinobu Maejima, A. Mizokawa, Daiki Kuwahara, S. Morishima","doi":"10.1109/ACPR.2013.187","DOIUrl":"https://doi.org/10.1109/ACPR.2013.187","url":null,"abstract":"We propose a facial aging simulator which can synthesize a photo realistic human aged-face image for criminal investigation. Our aging simulator is based on the patch-based facial texture reconstruction with a wrinkle aging pattern model. The advantage of our method is to synthesize an aged-face image with detailed skin texture such as spots and somberness of facial skin, as well as age-related facial wrinkles without blurs that are derived from lack of accurate pixel-wise alignments as in the linear combination model, while maintaining the identity of the original face.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129947844","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The generation of caricaturing portrait with exaggeration from real face image is one of the hot spots in the field of animation creation and digital multimedia entertainment. Based on improvement of traditional ASM facial feature points, this paper makes detailed definition of the human facial features and describes them in a way of proportion. Then, we propose a method which is based on the facial features and the relationship between them to generate exaggerated portrait from real face image. This method also introduces "contrast principle" while getting the exaggerated shape of face from two main aspects-facial form exaggeration and five senses organs exaggeration. At last, this method combines MLS image deformation method which is based on feature points to generate the exaggerated portrait of face image. Our experiments show that this method is practicable and can finally get results with good effect.
{"title":"A Method for Exaggerative Caricature Generation from Real Face Image","authors":"Chenglong Li, Z. Miao","doi":"10.1109/ACPR.2013.150","DOIUrl":"https://doi.org/10.1109/ACPR.2013.150","url":null,"abstract":"The generation of caricaturing portrait with exaggeration from real face image is one of the hot spots in the field of animation creation and digital multimedia entertainment. Based on improvement of traditional ASM facial feature points, this paper makes detailed definition of the human facial features and describes them in a way of proportion. Then, we propose a method which is based on the facial features and the relationship between them to generate exaggerated portrait from real face image. This method also introduces \"contrast principle\" while getting the exaggerated shape of face from two main aspects-facial form exaggeration and five senses organs exaggeration. At last, this method combines MLS image deformation method which is based on feature points to generate the exaggerated portrait of face image. Our experiments show that this method is practicable and can finally get results with good effect.","PeriodicalId":365633,"journal":{"name":"2013 2nd IAPR Asian Conference on Pattern Recognition","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2013-11-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125700603","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：