Pub Date : 2003-09-17DOI: 10.1109/ICIAP.2003.1234109
F. Tortorella
Support vector machines (SVM) are currently one of the classification systems most used in pattern recognition and data mining because of their accuracy and generalization capability. However, when dealing with very complex classification tasks where different errors bring different penalties, one should take into account the overall classification cost produced by the classifier more than its accuracy. It is thus necessary to provide some methods for tuning the SVM on the costs of the particular application. Depending on the characteristics of the cost matrix, this can be done during or after the learning phase of the classifier. In this paper we introduce two optimization schemes based on the two possible approaches and compare their performance on various data sets and kernels. The first experimental results show that both the proposed schemes are suitable for tuning SVM in cost-sensitive applications.
{"title":"An empirical comparison of in-learning and post-learning optimization schemes for tuning the support vector machines in cost-sensitive applications","authors":"F. Tortorella","doi":"10.1109/ICIAP.2003.1234109","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234109","url":null,"abstract":"Support vector machines (SVM) are currently one of the classification systems most used in pattern recognition and data mining because of their accuracy and generalization capability. However, when dealing with very complex classification tasks where different errors bring different penalties, one should take into account the overall classification cost produced by the classifier more than its accuracy. It is thus necessary to provide some methods for tuning the SVM on the costs of the particular application. Depending on the characteristics of the cost matrix, this can be done during or after the learning phase of the classifier. In this paper we introduce two optimization schemes based on the two possible approaches and compare their performance on various data sets and kernels. The first experimental results show that both the proposed schemes are suitable for tuning SVM in cost-sensitive applications.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"63 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131542190","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234086
S. Battiato, A. Bruna, A. Buemi, F. Naccari
In this paper we present a comparison between different approaches to CFA (colour filter array) image encoding. We show different performance offered by a new algorithm based on a vector quantization technique, JPEG-LS, a low complexity encoding standard and classical JPEG. We also show the effects of CFA image encoding on the colour reconstructed images by a typical image generation pipeline. A discussion about the computational complexity and memory requirement of the different encoding approaches is also presented.
{"title":"Coding techniques for CFA data images","authors":"S. Battiato, A. Bruna, A. Buemi, F. Naccari","doi":"10.1109/ICIAP.2003.1234086","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234086","url":null,"abstract":"In this paper we present a comparison between different approaches to CFA (colour filter array) image encoding. We show different performance offered by a new algorithm based on a vector quantization technique, JPEG-LS, a low complexity encoding standard and classical JPEG. We also show the effects of CFA image encoding on the colour reconstructed images by a typical image generation pipeline. A discussion about the computational complexity and memory requirement of the different encoding approaches is also presented.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"45 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132636791","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234020
L. Lombardi, A. Petrosino
We present an efficient and fully parallel 2D object recognition method based on the use of a multiscale tree representation of the object boundary and recursive learning of trees. Specifically, the object is represented by means of a tree where each node, corresponding to a boundary segment at some level of resolution, is characterized by a real vector containing curvature, length, and symmetry of the boundary segment, while the nodes are connected by arcs when segments at successive levels are spatially related. The recognition procedure is formulated as a training procedure made by recursive neural networks followed by a testing procedure over unknown tree structured patterns.
{"title":"Shape recognition by distributed recursive learning of multiscale trees","authors":"L. Lombardi, A. Petrosino","doi":"10.1109/ICIAP.2003.1234020","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234020","url":null,"abstract":"We present an efficient and fully parallel 2D object recognition method based on the use of a multiscale tree representation of the object boundary and recursive learning of trees. Specifically, the object is represented by means of a tree where each node, corresponding to a boundary segment at some level of resolution, is characterized by a real vector containing curvature, length, and symmetry of the boundary segment, while the nodes are connected by arcs when segments at successive levels are spatially related. The recognition procedure is formulated as a training procedure made by recursive neural networks followed by a testing procedure over unknown tree structured patterns.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133140909","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234042
W. Chojnacki, M. Brooks, A. Hengel, D. Gawley
Problems requiring accurate determination of parameters from image-based quantities arise often in computer vision. Two recent, independently developed frameworks for estimating such parameters are the FNS and HEIV schemes. Here it is shown that FNS (fundamental numerical scheme) and a core version of HEIV (heteroscedastic errors-in-variables) are essentially equivalent, solving a common underlying equation via different means. The analysis is driven by the search for a nondegenerate form of a certain generalised eigenvalue problem, and effectively leads to a new derivation of the relevant case of the HEIV algorithm. This work may be seen as an extension of previous efforts to rationalise and inter-relate a spectrum of estimators, including the renormalisation method of Kanatani and the normalised eight-point method of Hartley.
{"title":"FNS and HEIV: relating two vision parameter estimation frameworks","authors":"W. Chojnacki, M. Brooks, A. Hengel, D. Gawley","doi":"10.1109/ICIAP.2003.1234042","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234042","url":null,"abstract":"Problems requiring accurate determination of parameters from image-based quantities arise often in computer vision. Two recent, independently developed frameworks for estimating such parameters are the FNS and HEIV schemes. Here it is shown that FNS (fundamental numerical scheme) and a core version of HEIV (heteroscedastic errors-in-variables) are essentially equivalent, solving a common underlying equation via different means. The analysis is driven by the search for a nondegenerate form of a certain generalised eigenvalue problem, and effectively leads to a new derivation of the relevant case of the HEIV algorithm. This work may be seen as an extension of previous efforts to rationalise and inter-relate a spectrum of estimators, including the renormalisation method of Kanatani and the normalised eight-point method of Hartley.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"151 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116629103","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234094
Ikuhisa Mitsugami, N. Ukita, M. Kidode
We propose a new wearable system that can estimate the 3D position of a gazed point by measuring multiple binocular view lines. In principle, 3D measurement is possible by the triangulation of binocular view lines. However, it is difficult to measure these lines accurately with a device for eye tracking, because of errors caused by (1) difficulty in calibrating the device and (2) the limitation that a human cannot gaze very accurately at a distant point. Concerning (1), the accuracy of calibration can be improved by considering the optical properties of a camera in the device. To solve (2), we propose a stochastic algorithm that determines a gazed 3D position by integrating information of view lines observed at multiple head positions. We validated the effectiveness of the proposed algorithm experimentally.
{"title":"Estimation of 3D gazed position using view lines","authors":"Ikuhisa Mitsugami, N. Ukita, M. Kidode","doi":"10.1109/ICIAP.2003.1234094","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234094","url":null,"abstract":"We propose a new wearable system that can estimate the 3D position of a gazed point by measuring multiple binocular view lines. In principle, 3D measurement is possible by the triangulation of binocular view lines. However, it is difficult to measure these lines accurately with a device for eye tracking, because of errors caused by (1) difficulty in calibrating the device and (2) the limitation that a human cannot gaze very accurately at a distant point. Concerning (1), the accuracy of calibration can be improved by considering the optical properties of a camera in the device. To solve (2), we propose a stochastic algorithm that determines a gazed 3D position by integrating information of view lines observed at multiple head positions. We validated the effectiveness of the proposed algorithm experimentally.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114953352","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234078
B. Telle, M. Aldon, N. Ramdani
The paper deals with the problem of error estimation in 3D reconstruction. It shows how interval analysis can be used in this way for 3D vision applications. The description of an image point by an interval assumes an unknown but bounded localization. We present a new method based on interval analysis tools to propagate this bounded uncertainty. This way of computation can produce guaranteed results since a datum is not the most probabilistic value but an interval which contains the true value. We validate our method by computing a guaranteed model for a projective camera, and we achieve a guaranteed 3D reconstruction.
{"title":"Camera calibration and 3D reconstruction using interval analysis","authors":"B. Telle, M. Aldon, N. Ramdani","doi":"10.1109/ICIAP.2003.1234078","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234078","url":null,"abstract":"The paper deals with the problem of error estimation in 3D reconstruction. It shows how interval analysis can be used in this way for 3D vision applications. The description of an image point by an interval assumes an unknown but bounded localization. We present a new method based on interval analysis tools to propagate this bounded uncertainty. This way of computation can produce guaranteed results since a datum is not the most probabilistic value but an interval which contains the true value. We validate our method by computing a guaranteed model for a projective camera, and we achieve a guaranteed 3D reconstruction.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"73 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116350497","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234046
S. Bouchafa, B. Zavidovique
A new level-line registration technique is proposed for image transform estimation. This approach is robust towards contrast changes, does not require any estimate of the unknown transformation between images and tackles very challenging situations that usually lead to pairing ambiguities, such as repetitive patterns in the images. The registration itself is performed through an efficient level-line cumulative matching based on a multistage primitive election procedure. Each stage provides a coarse estimate of the transformation that the next stage gets to refine. Although we deal with similarity transforms (rotation, scale and translation), our approach can be easily adapted to more general transformations.
{"title":"Cumulative level-line matching for image registration","authors":"S. Bouchafa, B. Zavidovique","doi":"10.1109/ICIAP.2003.1234046","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234046","url":null,"abstract":"A new level-line registration technique is proposed for image transform estimation. This approach is robust towards contrast changes, does not require any estimate of the unknown transformation between images and tackles very challenging situations that usually lead to pairing ambiguities, such as repetitive patterns in the images. The registration itself is performed through an efficient level-line cumulative matching based on a multistage primitive election procedure. Each stage provides a coarse estimate of the transformation that the next stage gets to refine. Although we deal with similarity transforms (rotation, scale and translation), our approach can be easily adapted to more general transformations.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114888437","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234126
W. Clocksin, P. P. Fernando
We describe a method implemented for the recognition of Syriac handwriting from historical manuscripts. The Syriac language has been a neglected area for handwriting recognition research, yet is interesting because the preponderance of scribe-written manuscripts offers a challenging yet tractable medium for OCR research between the extremes of typewritten text and free handwriting. Like Arabic, Syriac is written in a cursive form from right-to-left, and letter shape depends on the position within the word. The method described does not need to find character strokes or contours. Both whole words and character shapes were used in recognition experiments. After segmentation using a novel probabilistic method, features of these shapes are found that tolerate variation in formation and image quality. Each shape is recognised individually using a discriminative support vector machine with 10-fold cross-validation. We describe experiments using a variety of segmentation methods and combinations of features on characters and words. Images from scribe-written historical manuscripts are used, and the recognition results are compared with those for images taken from clearer 19th century typeset documents. Recognition rates vary from 61-100%, depending on the algorithms used and the size and source of the data set.
{"title":"Towards automatic transcription of Syriac handwriting","authors":"W. Clocksin, P. P. Fernando","doi":"10.1109/ICIAP.2003.1234126","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234126","url":null,"abstract":"We describe a method implemented for the recognition of Syriac handwriting from historical manuscripts. The Syriac language has been a neglected area for handwriting recognition research, yet is interesting because the preponderance of scribe-written manuscripts offers a challenging yet tractable medium for OCR research between the extremes of typewritten text and free handwriting. Like Arabic, Syriac is written in a cursive form from right-to-left, and letter shape depends on the position within the word. The method described does not need to find character strokes or contours. Both whole words and character shapes were used in recognition experiments. After segmentation using a novel probabilistic method, features of these shapes are found that tolerate variation in formation and image quality. Each shape is recognised individually using a discriminative support vector machine with 10-fold cross-validation. We describe experiments using a variety of segmentation methods and combinations of features on characters and words. Images from scribe-written historical manuscripts are used, and the recognition results are compared with those for images taken from clearer 19th century typeset documents. Recognition rates vary from 61-100%, depending on the algorithms used and the size and source of the data set.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121144445","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234110
A. Barla, F. Odone, A. Verri
In this paper we present a statistical learning scheme for image classification based on a mixture of old fashioned ideas and state of the art learning tools. We represent input images through large dimensional and usually sparse histograms which, depending on the task, are either color histograms or co-occurrence matrices. Support vector machines are trained on these sparse inputs directly, to solve problems like indoor/outdoor classification and cityscape retrieval from image databases. The experimental results indicate that the use of a kernel function derived from the computer vision literature leads to better recognition results than off the shelf kernels. According to our findings, it appears that image classification problems can be addressed with no need of explicit feature extraction or dimensionality reduction stages. We argue that this might be used as the starting point for developing image classification systems which can be easily tuned to a number of different tasks.
{"title":"Old fashioned state-of-the-art image classification","authors":"A. Barla, F. Odone, A. Verri","doi":"10.1109/ICIAP.2003.1234110","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234110","url":null,"abstract":"In this paper we present a statistical learning scheme for image classification based on a mixture of old fashioned ideas and state of the art learning tools. We represent input images through large dimensional and usually sparse histograms which, depending on the task, are either color histograms or co-occurrence matrices. Support vector machines are trained on these sparse inputs directly, to solve problems like indoor/outdoor classification and cityscape retrieval from image databases. The experimental results indicate that the use of a kernel function derived from the computer vision literature leads to better recognition results than off the shelf kernels. According to our findings, it appears that image classification problems can be addressed with no need of explicit feature extraction or dimensionality reduction stages. We argue that this might be used as the starting point for developing image classification systems which can be easily tuned to a number of different tasks.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"44 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123866572","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 : 2003-09-17DOI: 10.1109/ICIAP.2003.1234098
Shih-Fu Chang
Today's mobile and wireless users access multimedia content from different types of networks and terminals. Content analysis plays a critical role in developing effective solutions in meeting unique resource constraints and user preferences in such usage environments. Specifically, content analysis is central to automatic discovery of syntactic-level summaries and generation of concise semantic-level summaries. Content analysis also provides a promising direction for finding optimal adaptation methods under various resource-utility constraints. The paper presents brief overviews of such emerging, fruitful areas and promising research directions.
{"title":"Content-based video summarization and adaptation for ubiquitous media access","authors":"Shih-Fu Chang","doi":"10.1109/ICIAP.2003.1234098","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234098","url":null,"abstract":"Today's mobile and wireless users access multimedia content from different types of networks and terminals. Content analysis plays a critical role in developing effective solutions in meeting unique resource constraints and user preferences in such usage environments. Specifically, content analysis is central to automatic discovery of syntactic-level summaries and generation of concise semantic-level summaries. Content analysis also provides a promising direction for finding optimal adaptation methods under various resource-utility constraints. The paper presents brief overviews of such emerging, fruitful areas and promising research directions.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"70 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2003-09-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124231282","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：