Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344000
Huixian Sun, Yu-hua Zhang, F. Luo
It is a difficult problem to describe the feature of wire rope texture with fault. The one-class classification method is adopted to description the feature of the faultless wire rope images. A method is presented to detect the surface defect of wire rope based the Support Vector Data Description (SVDD) method. The model selection and parameter optimize methods of SVDD are discussed thoroughly. Then, the bandwidth of Gauss kernel function is optimized to minimize the mean of false alarm rate in the experiment. The experiment is carried out to detect the surface fault of airplane control ropes with different diameters (4-6mm). The test of defect detection is carried out in 200 wire rope images, and the results indicate that the detecting accuracy is 93%. The method is valuable for detecting the surface local fault of aircraft control rope practically.
{"title":"Texture Defect Detection of Wire Rope Surface with Support Vector Data Description","authors":"Huixian Sun, Yu-hua Zhang, F. Luo","doi":"10.1109/CCPR.2009.5344000","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344000","url":null,"abstract":"It is a difficult problem to describe the feature of wire rope texture with fault. The one-class classification method is adopted to description the feature of the faultless wire rope images. A method is presented to detect the surface defect of wire rope based the Support Vector Data Description (SVDD) method. The model selection and parameter optimize methods of SVDD are discussed thoroughly. Then, the bandwidth of Gauss kernel function is optimized to minimize the mean of false alarm rate in the experiment. The experiment is carried out to detect the surface fault of airplane control ropes with different diameters (4-6mm). The test of defect detection is carried out in 200 wire rope images, and the results indicate that the detecting accuracy is 93%. The method is valuable for detecting the surface local fault of aircraft control rope practically.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116803430","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344141
Bin Chen, Bin Li, Zhisong Pan, AiMin Feng
Differed from the bounding hypersphere assumption in Support Vector Machine (SVM), Ellipsoidal Kernel Machine (EKM) adopts the compacter bounding ellipsoid assumption, and finds the separating plane inside the ellipsoid. It reduces the VC dimension in essence. However, EKM only applies in binary classification and does not work in outlier detection where generally only one class of samples existed. Thus, this paper proposes a method for outlier detection-One-class Ellipsoidal Machine and its kernel extension, which first finds a minimal ellipsoid enclosing all the input samples, and then finds the separating plane inside the ellipsoid by one-class SVM. Experiments on the artificial dataset and real datasets from UCI repository validate the effectiveness of the proposed method.
{"title":"One-Class Ellipsoidal Kernel Machine for Outlier Detection","authors":"Bin Chen, Bin Li, Zhisong Pan, AiMin Feng","doi":"10.1109/CCPR.2009.5344141","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344141","url":null,"abstract":"Differed from the bounding hypersphere assumption in Support Vector Machine (SVM), Ellipsoidal Kernel Machine (EKM) adopts the compacter bounding ellipsoid assumption, and finds the separating plane inside the ellipsoid. It reduces the VC dimension in essence. However, EKM only applies in binary classification and does not work in outlier detection where generally only one class of samples existed. Thus, this paper proposes a method for outlier detection-One-class Ellipsoidal Machine and its kernel extension, which first finds a minimal ellipsoid enclosing all the input samples, and then finds the separating plane inside the ellipsoid by one-class SVM. Experiments on the artificial dataset and real datasets from UCI repository validate the effectiveness of the proposed method.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"54 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127381071","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344081
Gaopeng Zhao, Y. Bo, Ming Lv
Aimed at the problem of dim small target detection in infrared image which is characterized by low SNR and the serious background and noise disturbance, a new detection method was proposed based on the nonsubsampled contourlet transform. Firstly, the nonsubsampled contourlet transform was employed to decompose the source image. Considering the different energy distribution characteristics of the target, background and noise, the energy image was constructed by energy cross correlation operation. Then, a few of candidate targets were segmented through adaptive threshold. Finally, the target was confirmed by utilizing the continuity and consistency of target energy intensity and target movement in image sequence. The experimental results showed that the method is effectively and the dim small target can be detected accurately in infrared image sequence.
{"title":"Dim Small Target Detection Method Based on Nonsubsampled Contourlet Transform in Infrared Image","authors":"Gaopeng Zhao, Y. Bo, Ming Lv","doi":"10.1109/CCPR.2009.5344081","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344081","url":null,"abstract":"Aimed at the problem of dim small target detection in infrared image which is characterized by low SNR and the serious background and noise disturbance, a new detection method was proposed based on the nonsubsampled contourlet transform. Firstly, the nonsubsampled contourlet transform was employed to decompose the source image. Considering the different energy distribution characteristics of the target, background and noise, the energy image was constructed by energy cross correlation operation. Then, a few of candidate targets were segmented through adaptive threshold. Finally, the target was confirmed by utilizing the continuity and consistency of target energy intensity and target movement in image sequence. The experimental results showed that the method is effectively and the dim small target can be detected accurately in infrared image sequence.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"61 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126657266","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344138
Jianxin Huang, Yanyun Qu, Cui-hua Li, Miao-Jun Hu
Object categorization has become active recently in the field of pattern recognition. There are two main factors which affect the performance of classification. One is the representation of images, and the other is the design of classifier. The representation of images based on bag-of-word(BOW) has become a popular method because of its simpleness and high efficiency. This paper aims to compare some state-of-the-art classifiers used in object categorization based on the BOW technology. In the dataset of Xerox7 and CalTech6, we compare the performance of five classifiers which are SVM, Maximum Entropy, Naive Bayes, Adaboost and Random Forests. The result of experiments show that SVM and Maximum Entropy have better performance than others.
{"title":"The Comparison of Classifiers for Object Categorization Based on Bag-of-Word Technology","authors":"Jianxin Huang, Yanyun Qu, Cui-hua Li, Miao-Jun Hu","doi":"10.1109/CCPR.2009.5344138","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344138","url":null,"abstract":"Object categorization has become active recently in the field of pattern recognition. There are two main factors which affect the performance of classification. One is the representation of images, and the other is the design of classifier. The representation of images based on bag-of-word(BOW) has become a popular method because of its simpleness and high efficiency. This paper aims to compare some state-of-the-art classifiers used in object categorization based on the BOW technology. In the dataset of Xerox7 and CalTech6, we compare the performance of five classifiers which are SVM, Maximum Entropy, Naive Bayes, Adaboost and Random Forests. The result of experiments show that SVM and Maximum Entropy have better performance than others.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124824652","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344102
Guo Cao, Yuan Mei, Quansen Sun
In this paper, we propose an image partitioning method using level set evolution for an arbitrary number of regions and embark on the concept of using one level set function for each region. The energy functional of each level set uses shifted Heaviside functions to obtain a stationary global minimum, which makes the proposed algorithm invariant to the level set initialization. In addition, unlike most of the previous works, the curve evolution partial differential equations for different level set equations are decoupled by applying the min operator and the proposed algorithm allows the effective number of regions to vary during the evolving process. Each region of class evolves according to its features and competes with the neighbor regions in order to get a partition. Generally, the proposed algorithm is fast, easy to implement, and not sensitive to the choice of initial conditions. Results are shown on both synthetic and real images.
{"title":"Region Competition Based Active Contour for Image Partitioning","authors":"Guo Cao, Yuan Mei, Quansen Sun","doi":"10.1109/CCPR.2009.5344102","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344102","url":null,"abstract":"In this paper, we propose an image partitioning method using level set evolution for an arbitrary number of regions and embark on the concept of using one level set function for each region. The energy functional of each level set uses shifted Heaviside functions to obtain a stationary global minimum, which makes the proposed algorithm invariant to the level set initialization. In addition, unlike most of the previous works, the curve evolution partial differential equations for different level set equations are decoupled by applying the min operator and the proposed algorithm allows the effective number of regions to vary during the evolving process. Each region of class evolves according to its features and competes with the neighbor regions in order to get a partition. Generally, the proposed algorithm is fast, easy to implement, and not sensitive to the choice of initial conditions. Results are shown on both synthetic and real images.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"83 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122634805","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344042
Jianming Wang, R. Dou, Zhijie Yan, Zhongwei Wang, Zhong Zhang
A method to measure the emotion strength in speech was proposed in the paper. Firstly, it was assumed that neutral speech doesn't contain any emotion and then the Speech Emotion Shifting Assumption was proposed. A measure function was determined by maximizing the Speech Emotion Shifting Assumption, by which quantification values of speech samples were calculated. A speech database with different emotion strength levels was built in the order of verifying the emotion strength analysis method. The experimental results showed that the method is reasonable for measuring the emotion strength in speech.
{"title":"Exploration of Analyzing Emotion Strength in Speech Signal","authors":"Jianming Wang, R. Dou, Zhijie Yan, Zhongwei Wang, Zhong Zhang","doi":"10.1109/CCPR.2009.5344042","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344042","url":null,"abstract":"A method to measure the emotion strength in speech was proposed in the paper. Firstly, it was assumed that neutral speech doesn't contain any emotion and then the Speech Emotion Shifting Assumption was proposed. A measure function was determined by maximizing the Speech Emotion Shifting Assumption, by which quantification values of speech samples were calculated. A speech database with different emotion strength levels was built in the order of verifying the emotion strength analysis method. The experimental results showed that the method is reasonable for measuring the emotion strength in speech.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"184 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123521483","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344001
Peng Hu, Yangyu Luo, Chengrong Li
The recognition of Chinese chess through computer vision includes chess detection and its character recognition, and the process should be fast and robust. In this paper, Robert operator is used to get the edge information of chess, then detection, localization and segmentation of the chess are realized through mathematical morphology and template circle method. A new algorithm based on projection histogram of polar coordinates image and Fast Fourier Transform is introduced to extract the rotation-invariant feature of chess characters. Experiments show that the algorithm can accurately recognize all chesses within 300 ms, and is robust to any rotation.
{"title":"Chinese Chess Recognition Based on Projection Histogram of Polar Coordinates Image and FFT","authors":"Peng Hu, Yangyu Luo, Chengrong Li","doi":"10.1109/CCPR.2009.5344001","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344001","url":null,"abstract":"The recognition of Chinese chess through computer vision includes chess detection and its character recognition, and the process should be fast and robust. In this paper, Robert operator is used to get the edge information of chess, then detection, localization and segmentation of the chess are realized through mathematical morphology and template circle method. A new algorithm based on projection histogram of polar coordinates image and Fast Fourier Transform is introduced to extract the rotation-invariant feature of chess characters. Experiments show that the algorithm can accurately recognize all chesses within 300 ms, and is robust to any rotation.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131375625","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5343953
Jinsu Jo, Jihyun Lee, Yillbyung Lee
In this paper we propose a stroke recognition method for online handwritten Hangul recognition system. The proposed system extracts a distance-dependent curvature from two-dimensional original stroke data and achieves elastic matching between distance-dependent curvatures of reference and test characters. Elastic curvature matching has lower computational requirement than existing 2D-to-2D elastic matching. Each recognized stroke from the elastic curvature matching is converted into a Hangul syllable and additional position information is added to improve performance of the recognizer in this process.
{"title":"Stroke-Based Online Hangul/Korean Character Recognition","authors":"Jinsu Jo, Jihyun Lee, Yillbyung Lee","doi":"10.1109/CCPR.2009.5343953","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5343953","url":null,"abstract":"In this paper we propose a stroke recognition method for online handwritten Hangul recognition system. The proposed system extracts a distance-dependent curvature from two-dimensional original stroke data and achieves elastic matching between distance-dependent curvatures of reference and test characters. Elastic curvature matching has lower computational requirement than existing 2D-to-2D elastic matching. Each recognized stroke from the elastic curvature matching is converted into a Hangul syllable and additional position information is added to improve performance of the recognizer in this process.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131622426","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344087
Yong Hu, Chunxia Zhao
In the field of unsupervised texture classification, a combination of various families of methods was usually used for better classification results. However, the existing methods are usually used for specific application and evaluated with fixed window size. In this literature, we propose an effort to combine multi-scale features for unsupervised texture classification. The local binary pattern (LBP) is used for detecting micro textured structures. As for large scale texture information, Haralick features extracted from gray level co-occurrence matrix (GLCM) are adopted. In order to determine the optimal window size, each method is evaluated with different window sizes. By combining the information provided by multi-scale features for classification, the proposed method achieved higher classification rate than each single method evaluated over fixed window size. Experimental results confirmed the usefulness of this combination.
{"title":"Unsupervised Texture Classification by Combining Multi-Scale Features and K-Means Classifier","authors":"Yong Hu, Chunxia Zhao","doi":"10.1109/CCPR.2009.5344087","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344087","url":null,"abstract":"In the field of unsupervised texture classification, a combination of various families of methods was usually used for better classification results. However, the existing methods are usually used for specific application and evaluated with fixed window size. In this literature, we propose an effort to combine multi-scale features for unsupervised texture classification. The local binary pattern (LBP) is used for detecting micro textured structures. As for large scale texture information, Haralick features extracted from gray level co-occurrence matrix (GLCM) are adopted. In order to determine the optimal window size, each method is evaluated with different window sizes. By combining the information provided by multi-scale features for classification, the proposed method achieved higher classification rate than each single method evaluated over fixed window size. Experimental results confirmed the usefulness of this combination.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130443025","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}
Pub Date : 2009-12-04DOI: 10.1109/CCPR.2009.5344134
Mingming Sun, Chuancai Liu, Jingyu Yang
Various of manifold learning methods have been proposed to capture the intrinsic characteristic of nonlinear data. However, when con- fronting highly nonlinear data sets, existing algorithms may fail to discover the correct inner structure of data sets. In this paper, we proposed a new locality-based manifold learning method - Neighborhood Balance Embedding. The proposed method share the same 'neighborhood preserving' property with other manifold learning methods, however, it describe the local structure in a different way, which makes each neighborhood like a s rigid balls, thus prevents the overlapping phenomenon which often happens when coping with highly nonlinear data. Experimental results on the data sets with high nonlinearity show good performances of the proposed method. However, when dealing with the highly nonlinear data sets, the existing methods may fail to find reasonable solution. These failures are derived from several aspects. First, many methods may achieve optimal results only for a limited kinds of mani- folds (for example, global or local isometric manifold), such as LTSA, MVU and ISOMap. However, highly nonlinear data sets may lie on manifolds that may not satisfy such constraints, thus the methods may not capture the intrinsic character of the data set. Second, the local structures constructed by many meth- ods such as LLE and LEM do not prevent the features of the samples in a neighborhood from being excessively close from each other. This may result in overlapping phenomenon, that is, mapping the removed samples near into a neighborhood in the feature space. This phenomenon results in a misleading com- prehension of the data set, and a large bias from the reconstruc- tion manifold to the data sets. In this paper, we proposed a new embedding method for manifold learning - Neighborhood Balance Embedding. The method preserves the neighborhood relationship between the samples. However, when finding the local structure of neigh- borhoods, the method makes a balance among the distances be- tween the features in a neighborhood, that is, when two samples are close, their features are required to be little further away from each other; when two samples are far from each other (still in a neighborhood), their features are required to be little closer from each other. In this manner, the neighborhoods of the fea- tures are constructed like a set of rigid balls, thus prevents the overlapping phenomenon. Experimental results show the suc- cesses of the proposed methods on some data sets on which many existing manifold learning methods would fail.
{"title":"Neighborhood Balance Embedding for Unsupervised Dimensionality Reduction","authors":"Mingming Sun, Chuancai Liu, Jingyu Yang","doi":"10.1109/CCPR.2009.5344134","DOIUrl":"https://doi.org/10.1109/CCPR.2009.5344134","url":null,"abstract":"Various of manifold learning methods have been proposed to capture the intrinsic characteristic of nonlinear data. However, when con- fronting highly nonlinear data sets, existing algorithms may fail to discover the correct inner structure of data sets. In this paper, we proposed a new locality-based manifold learning method - Neighborhood Balance Embedding. The proposed method share the same 'neighborhood preserving' property with other manifold learning methods, however, it describe the local structure in a different way, which makes each neighborhood like a s rigid balls, thus prevents the overlapping phenomenon which often happens when coping with highly nonlinear data. Experimental results on the data sets with high nonlinearity show good performances of the proposed method. However, when dealing with the highly nonlinear data sets, the existing methods may fail to find reasonable solution. These failures are derived from several aspects. First, many methods may achieve optimal results only for a limited kinds of mani- folds (for example, global or local isometric manifold), such as LTSA, MVU and ISOMap. However, highly nonlinear data sets may lie on manifolds that may not satisfy such constraints, thus the methods may not capture the intrinsic character of the data set. Second, the local structures constructed by many meth- ods such as LLE and LEM do not prevent the features of the samples in a neighborhood from being excessively close from each other. This may result in overlapping phenomenon, that is, mapping the removed samples near into a neighborhood in the feature space. This phenomenon results in a misleading com- prehension of the data set, and a large bias from the reconstruc- tion manifold to the data sets. In this paper, we proposed a new embedding method for manifold learning - Neighborhood Balance Embedding. The method preserves the neighborhood relationship between the samples. However, when finding the local structure of neigh- borhoods, the method makes a balance among the distances be- tween the features in a neighborhood, that is, when two samples are close, their features are required to be little further away from each other; when two samples are far from each other (still in a neighborhood), their features are required to be little closer from each other. In this manner, the neighborhoods of the fea- tures are constructed like a set of rigid balls, thus prevents the overlapping phenomenon. Experimental results show the suc- cesses of the proposed methods on some data sets on which many existing manifold learning methods would fail.","PeriodicalId":354468,"journal":{"name":"2009 Chinese Conference on Pattern Recognition","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2009-12-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116579844","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：