Pub Date : 2002-12-10DOI: 10.1109/ICPR.2002.1048410
C. Chennubhotla, A. Jepson, J. Midgley
We achieve two goals in this paper: (1) to build a novel appearance-based object representation that takes into account variations in contrast often found in training images; (2) to develop a robust appearance-based detection scheme that can handle outliers such as occlusion and structured noise. To build the representation, we decompose the input ensemble into two subspaces: a principal subspace (within-subspace) and its orthogonal complement (out-of-subspace). Before computing the principal subspace, we remove any dependency on contrast that the training set might exhibit. To account for pixel outliers in test images, we model the residual signal in the out-of-subspace by a probabilistic mixture model of an inlier distribution and a uniform outlier distribution. The mixture model, in turn, facilitates the robust estimation of the within-subspace coefficients. We show our methodology leads to an effective classifier for separating images of eyes from non-eyes extracted from the FERET dataset.
{"title":"Robust contrast-invariant eigen detection","authors":"C. Chennubhotla, A. Jepson, J. Midgley","doi":"10.1109/ICPR.2002.1048410","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048410","url":null,"abstract":"We achieve two goals in this paper: (1) to build a novel appearance-based object representation that takes into account variations in contrast often found in training images; (2) to develop a robust appearance-based detection scheme that can handle outliers such as occlusion and structured noise. To build the representation, we decompose the input ensemble into two subspaces: a principal subspace (within-subspace) and its orthogonal complement (out-of-subspace). Before computing the principal subspace, we remove any dependency on contrast that the training set might exhibit. To account for pixel outliers in test images, we model the residual signal in the out-of-subspace by a probabilistic mixture model of an inlier distribution and a uniform outlier distribution. The mixture model, in turn, facilitates the robust estimation of the within-subspace coefficients. We show our methodology leads to an effective classifier for separating images of eyes from non-eyes extracted from the FERET dataset.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125072537","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048445
O. Ecabert, J. Thiran
This paper presents a novel variational image segmentation technique that unifies both geodesic active contours and geodesic active regions. The originality of the method is the automatic and dynamic global weighting of the respective local equations of motion. A new stopping function for the geodesic active contours is also introduced, which proves to have a better behavior in the vicinity of the object boundaries. Instead of minimizing the standard energy functional, we use a normalized version, which strongly reduces the shortening effect, improving thus the coupling with the region model. Results and method effectiveness are shown on real and medical images.
{"title":"Variational image segmentation by unifying region and boundary information","authors":"O. Ecabert, J. Thiran","doi":"10.1109/ICPR.2002.1048445","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048445","url":null,"abstract":"This paper presents a novel variational image segmentation technique that unifies both geodesic active contours and geodesic active regions. The originality of the method is the automatic and dynamic global weighting of the respective local equations of motion. A new stopping function for the geodesic active contours is also introduced, which proves to have a better behavior in the vicinity of the object boundaries. Instead of minimizing the standard energy functional, we use a normalized version, which strongly reduces the shortening effect, improving thus the coupling with the region model. Results and method effectiveness are shown on real and medical images.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126043547","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048367
Changjiang Yang, R. Duraiswami, L. Davis
Computer vision requires the solution of many ill-posed problems such as optical flow, structure from motion, shape from shading, surface reconstruction, image restoration and edge detection. Regularization is a popular method to solve ill-posed problems, in which the solution is sought by minimization of a sum of two weighted terms, one measuring the error arising from the ill-posed model, the other indicating the distance between the solution and some class of solutions chosen on the basis of prior knowledge (smoothness, or other prior information). One of important issues in regularization is choosing optimal weight (or regularization parameter). Existing methods for choosing regularization parameters either require the prior information on noise in the data, or are heuristic graphical methods. We apply a method for choosing near-optimal regularization parameters by approximately minimizing the distance between the true solution and the family of regularized solutions. We demonstrate the effectiveness of this approach for the regularization on two examples: edge detection and image restoration.
{"title":"Near-optimal regularization parameters for applications in computer vision","authors":"Changjiang Yang, R. Duraiswami, L. Davis","doi":"10.1109/ICPR.2002.1048367","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048367","url":null,"abstract":"Computer vision requires the solution of many ill-posed problems such as optical flow, structure from motion, shape from shading, surface reconstruction, image restoration and edge detection. Regularization is a popular method to solve ill-posed problems, in which the solution is sought by minimization of a sum of two weighted terms, one measuring the error arising from the ill-posed model, the other indicating the distance between the solution and some class of solutions chosen on the basis of prior knowledge (smoothness, or other prior information). One of important issues in regularization is choosing optimal weight (or regularization parameter). Existing methods for choosing regularization parameters either require the prior information on noise in the data, or are heuristic graphical methods. We apply a method for choosing near-optimal regularization parameters by approximately minimizing the distance between the true solution and the family of regularized solutions. We demonstrate the effectiveness of this approach for the regularization on two examples: edge detection and image restoration.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"46 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125381939","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048275
Puneet Gupta, D. Doermann, D. DeMenthon
Often in pattern classification problems, one tries to extract a large number of features and base the classifier decision on as much information as possible. This yields an array of features that are 'potentially' useful. Most of the time however, large feature sets are sub-optimal in describing the samples since they tend to over-represent the data and model noise along with the useful information in the data. Selecting relevant features from the available set of features is, therefore, a challenging task. In this paper, we present an innovative feature selection algorithm called Smart Beam Search (SBS), which is used with a support vector machine (SVM) based classifier for automatic defect classification. This feature selection approach not only reduces the dimensionality of the feature space substantially, but also improves the classifier performance.
{"title":"Beam search for feature selection in automatic SVM defect classification","authors":"Puneet Gupta, D. Doermann, D. DeMenthon","doi":"10.1109/ICPR.2002.1048275","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048275","url":null,"abstract":"Often in pattern classification problems, one tries to extract a large number of features and base the classifier decision on as much information as possible. This yields an array of features that are 'potentially' useful. Most of the time however, large feature sets are sub-optimal in describing the samples since they tend to over-represent the data and model noise along with the useful information in the data. Selecting relevant features from the available set of features is, therefore, a challenging task. In this paper, we present an innovative feature selection algorithm called Smart Beam Search (SBS), which is used with a support vector machine (SVM) based classifier for automatic defect classification. This feature selection approach not only reduces the dimensionality of the feature space substantially, but also improves the classifier performance.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"12 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126714665","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048470
S. Bhagavathy, S. Newsam, B. S. Manjunath
We propose a canonical model for object classes in aerial images. This model is motivated by the observation that geographic regions of interest are characterized by collections of texture motifs corresponding to the geographic processes that generate them. We show that this model is effective in learning the common texture themes, or motifs, of the object classes.
{"title":"Modeling object classes in aerial images using texture motifs","authors":"S. Bhagavathy, S. Newsam, B. S. Manjunath","doi":"10.1109/ICPR.2002.1048470","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048470","url":null,"abstract":"We propose a canonical model for object classes in aerial images. This model is motivated by the observation that geographic regions of interest are characterized by collections of texture motifs corresponding to the geographic processes that generate them. We show that this model is effective in learning the common texture themes, or motifs, of the object classes.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122421228","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048458
M. L. Kherfi, D. Ziou, A. Bernardi
In this paper, we address some issues related to the combination of positive and negative examples to perform more efficient image retrieval. We analyze the relevance of negative example and how it can be interpreted. Then we propose a new relevance feedback model that integrates both positive and negative examples. First, a query is formulated using positive example, then negative example is used to refine the system's response. Mathematically, relevance feedback is formulated as an optimization of intra and inter variances of positive and negative examples.
{"title":"Learning from negative example in relevance feedback for content-based image retrieval","authors":"M. L. Kherfi, D. Ziou, A. Bernardi","doi":"10.1109/ICPR.2002.1048458","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048458","url":null,"abstract":"In this paper, we address some issues related to the combination of positive and negative examples to perform more efficient image retrieval. We analyze the relevance of negative example and how it can be interpreted. Then we propose a new relevance feedback model that integrates both positive and negative examples. First, a query is formulated using positive example, then negative example is used to refine the system's response. Mathematically, relevance feedback is formulated as an optimization of intra and inter variances of positive and negative examples.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122737238","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048269
V. Ryazanov, Victor A. Vorontchikhin
The proposed method for automatic knowledge extraction from large-scale data is based on the idea of analysing neighborhoods of "supporting" objects and construction of data covered by sets of hyper parallelepipeds. A simple procedure to choose the supporting objects is applied. Knowledge extraction (logical regularities search) is based on the solution of special discrete linear optimization tasks associated with supporting objects. Two practical tasks are considered for method illustration.
{"title":"Discrete approach for automatic knowledge extraction from precedent large-scale data, and classification","authors":"V. Ryazanov, Victor A. Vorontchikhin","doi":"10.1109/ICPR.2002.1048269","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048269","url":null,"abstract":"The proposed method for automatic knowledge extraction from large-scale data is based on the idea of analysing neighborhoods of \"supporting\" objects and construction of data covered by sets of hyper parallelepipeds. A simple procedure to choose the supporting objects is applied. Knowledge extraction (logical regularities search) is based on the solution of special discrete linear optimization tasks associated with supporting objects. Two practical tasks are considered for method illustration.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122934552","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048266
B. Schiele
Combining multiple classifiers promises to increase performance and robustness of a classification task. Currently, the understanding which combination scheme should be used and the ability to quantify the expected benefit is inadequate. This paper attempts to quantify the performance and robustness gain for different combination schemes and for two classifier types. The results indicate that the combination of a small number of classifiers may already result in a substantial performance gain. Also, the increase in robustness can be substantial by combining an adequate number of classifiers.
{"title":"How many classifiers do I need?","authors":"B. Schiele","doi":"10.1109/ICPR.2002.1048266","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048266","url":null,"abstract":"Combining multiple classifiers promises to increase performance and robustness of a classification task. Currently, the understanding which combination scheme should be used and the ability to quantify the expected benefit is inadequate. This paper attempts to quantify the performance and robustness gain for different combination schemes and for two classifier types. The results indicate that the combination of a small number of classifiers may already result in a substantial performance gain. Also, the increase in robustness can be substantial by combining an adequate number of classifiers.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"57 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122971942","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048395
M. Lalonde, L. Gagnon
This paper proposes two modifications to the geometrically deformable template model. First, the optimization stage originally based on simulated annealing is replaced with a meta-heuristic called Variable Neighborhood Search that treats simulated annealing as a local search tool. Second, an affine deformation energy is introduced to improve the quality of the search. An example of optic disc segmentation in an ophthalmic image is given.
{"title":"Variable neighborhood search for geometrically deformable templates","authors":"M. Lalonde, L. Gagnon","doi":"10.1109/ICPR.2002.1048395","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048395","url":null,"abstract":"This paper proposes two modifications to the geometrically deformable template model. First, the optimization stage originally based on simulated annealing is replaced with a meta-heuristic called Variable Neighborhood Search that treats simulated annealing as a local search tool. Second, an affine deformation energy is introduced to improve the quality of the search. An example of optic disc segmentation in an ophthalmic image is given.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"105 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128259427","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 : 2002-12-10DOI: 10.1109/ICPR.2002.1048332
Britta Bauer, K. Kraiss
This paper deals with the automatic recognition of German signs. The statistical approach is based on the Bayes decision rule for minimum error rate. Following speech recognition system designs, which are in general based on phonemes, here the idea of an automatic sign language recognition system using subunits rather than models for whole signs is outlined. The advantage of such a system will be a future reduction of necessary training material. Furthermore, a simplified enlargement of the existing vocabulary is expected, as new signs can be added to the vocabulary database without re-training the existing hidden Markov models (HMMs) for subunits. Since it is difficult to define subunits for sign language, this approach employs totally self-organized subunits. In first experiences a recognition accuracy of 92,5% was achieved for 100 signs, which were previously trained. For 50 new signs an accuracy of 81% was achieved without retraining of subunit-HMMs.
{"title":"Video-based sign recognition using self-organizing subunits","authors":"Britta Bauer, K. Kraiss","doi":"10.1109/ICPR.2002.1048332","DOIUrl":"https://doi.org/10.1109/ICPR.2002.1048332","url":null,"abstract":"This paper deals with the automatic recognition of German signs. The statistical approach is based on the Bayes decision rule for minimum error rate. Following speech recognition system designs, which are in general based on phonemes, here the idea of an automatic sign language recognition system using subunits rather than models for whole signs is outlined. The advantage of such a system will be a future reduction of necessary training material. Furthermore, a simplified enlargement of the existing vocabulary is expected, as new signs can be added to the vocabulary database without re-training the existing hidden Markov models (HMMs) for subunits. Since it is difficult to define subunits for sign language, this approach employs totally self-organized subunits. In first experiences a recognition accuracy of 92,5% was achieved for 100 signs, which were previously trained. For 50 new signs an accuracy of 81% was achieved without retraining of subunit-HMMs.","PeriodicalId":159502,"journal":{"name":"Object recognition supported by user interaction for service robots","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2002-12-10","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128666859","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：