Pub Date : 2003-09-17DOI: 10.1109/ICIAP.2003.1234025
K. Hotta
The paper presents a view-invariant face detection method based on local PCA cells. In order to extract the general features of faces at each view and position, Gabor filters and local PCA are used. Local PCA cells specialized to each view and position are made by applying a Gaussian to the outputs of the local PCA of Gabor features. By applying the Gaussian, only the local PCA cells which are a similar view to an input give large values. This decreases the bad influence of the local PCA cells of other views. As a result, only one classifier can treat multi-view faces well by integrating the outputs of local PCA cells. It is confirmed that the proposed method can detect multi-view faces. Generalization ability is improved by selecting the local PCA cells using a reconstruction error of local PCA.
{"title":"View-invariant face detection method based on local PCA cells","authors":"K. Hotta","doi":"10.1109/ICIAP.2003.1234025","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234025","url":null,"abstract":"The paper presents a view-invariant face detection method based on local PCA cells. In order to extract the general features of faces at each view and position, Gabor filters and local PCA are used. Local PCA cells specialized to each view and position are made by applying a Gaussian to the outputs of the local PCA of Gabor features. By applying the Gaussian, only the local PCA cells which are a similar view to an input give large values. This decreases the bad influence of the local PCA cells of other views. As a result, only one classifier can treat multi-view faces well by integrating the outputs of local PCA cells. It is confirmed that the proposed method can detect multi-view faces. Generalization ability is improved by selecting the local PCA cells using a reconstruction error of local PCA.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"25 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":"130202030","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.2003.1234121
L. Cordella, P. Foggia, Carlo Sansone, M. Vento
The paper presents a real-time speaker identification system based on the analysis of the audio track of a video stream. The system has been employed in the context of automatic video segmentation. It uses features evaluated in both the time and frequency domains. Their combined use significantly improve the performance of the system. Experiments have been carried on a database extracted from over one hour of television news, including 10 speakers. The obtained results confirm the effectiveness of the approach, showing an error rate less then 1% when the time interval used for identifying a speaker is about 1.5 seconds.
{"title":"A real-time text-independent speaker identification system","authors":"L. Cordella, P. Foggia, Carlo Sansone, M. Vento","doi":"10.1109/ICIAP.2003.1234121","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234121","url":null,"abstract":"The paper presents a real-time speaker identification system based on the analysis of the audio track of a video stream. The system has been employed in the context of automatic video segmentation. It uses features evaluated in both the time and frequency domains. Their combined use significantly improve the performance of the system. Experiments have been carried on a database extracted from over one hour of television news, including 10 speakers. The obtained results confirm the effectiveness of the approach, showing an error rate less then 1% when the time interval used for identifying a speaker is about 1.5 seconds.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"12 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":"122138926","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.2003.1234107
M. R. Hamid, Aijaz A. Baloch, A. Bilal, Nauman Zaffar
This paper presents a new technique to segment objects of interest from cluttered background with varying edge densities and illumination conditions from gray scale imagery. An optimal background model is generated and an index of disparity of the objects from this model is computed. This index estimates the disparity, both in terms of edge densities and edge orientation. We introduce feature based conditional morphology to process the representations that are most likely to belong to the object of interest and obtain a distilled edge map. These edges are linked using N/sup th/ order interpolation to get the final outline of the object. We compare our approach with 9 contemporary background subtraction algorithms (Toyama et al. (1999)). Our approach shows significant performance advantages and uses only the gray scale images, while the other approaches also need the color images for their algorithms. A comparison with the conventional morphological techniques is also made to highlight the advantages of our algorithms.
提出了一种从灰度图像中具有不同边缘密度和光照条件的杂乱背景中分割感兴趣目标的新技术。生成最优背景模型,并以此模型计算出目标的视差指数。该指数从边缘密度和边缘方向两方面来估计差异。我们引入基于特征的条件形态学来处理最有可能属于感兴趣对象的表示,并获得一个蒸馏的边缘映射。使用N/sup /阶插值将这些边连接起来,以获得对象的最终轮廓。我们将我们的方法与9种当代背景减法算法(Toyama et al.(1999))进行了比较。我们的方法显示出明显的性能优势,并且只使用灰度图像,而其他方法的算法也需要彩色图像。并与传统形态学技术进行了比较,以突出我们的算法的优势。
{"title":"Object segmentation using feature based conditional morphology","authors":"M. R. Hamid, Aijaz A. Baloch, A. Bilal, Nauman Zaffar","doi":"10.1109/ICIAP.2003.1234107","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234107","url":null,"abstract":"This paper presents a new technique to segment objects of interest from cluttered background with varying edge densities and illumination conditions from gray scale imagery. An optimal background model is generated and an index of disparity of the objects from this model is computed. This index estimates the disparity, both in terms of edge densities and edge orientation. We introduce feature based conditional morphology to process the representations that are most likely to belong to the object of interest and obtain a distilled edge map. These edges are linked using N/sup th/ order interpolation to get the final outline of the object. We compare our approach with 9 contemporary background subtraction algorithms (Toyama et al. (1999)). Our approach shows significant performance advantages and uses only the gray scale images, while the other approaches also need the color images for their algorithms. A comparison with the conventional morphological techniques is also made to highlight the advantages of our algorithms.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"29 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":"117275277","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.2003.1234128
S. Impedovo, G. Pirlo
The paper presents a new technique for the design of class-oriented recognizer. For each recognizer, a generic technique is used to determine, in an optimal way, the weights to balance the local decisions obtained from the analysis by parts of the patterns of the specific class. The experimental results, that have been obtained in the field of handwritten numeral and character recognition, demonstrate the superiority of the new technique with respect to other traditional approaches.
{"title":"Class-oriented recognizer design by weighting local decisions","authors":"S. Impedovo, G. Pirlo","doi":"10.1109/ICIAP.2003.1234128","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234128","url":null,"abstract":"The paper presents a new technique for the design of class-oriented recognizer. For each recognizer, a generic technique is used to determine, in an optimal way, the weights to balance the local decisions obtained from the analysis by parts of the patterns of the specific class. The experimental results, that have been obtained in the field of handwritten numeral and character recognition, demonstrate the superiority of the new technique with respect to other traditional approaches.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"50 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":"126277944","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.2003.1234039
H. Eng, A. H. Kam, Junxian Wang, W. Yau, Lijuan Jiang
Many deployed systems for human motion tracking and detection are found inadequate when applied on hostile outdoor environments. This paper provides insights into this problem by developing an outdoor aquatic surveillance system, which detects swimmers within the hostile environment of an outdoor public swimming pool. A novel block-based background model and thresholding-with-hysteresis methodology is proposed to extract swimmers amid reflections, ripples, splashes and lighting changes. The problem of partial occlusion between swimmers is resolved based on a proposed Markov random field framework. The algorithm has been incorporated into a live system with robust results for different challenging outdoor pool conditions.
{"title":"Human detection and tracking within hostile aquatic environments","authors":"H. Eng, A. H. Kam, Junxian Wang, W. Yau, Lijuan Jiang","doi":"10.1109/ICIAP.2003.1234039","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234039","url":null,"abstract":"Many deployed systems for human motion tracking and detection are found inadequate when applied on hostile outdoor environments. This paper provides insights into this problem by developing an outdoor aquatic surveillance system, which detects swimmers within the hostile environment of an outdoor public swimming pool. A novel block-based background model and thresholding-with-hysteresis methodology is proposed to extract swimmers amid reflections, ripples, splashes and lighting changes. The problem of partial occlusion between swimmers is resolved based on a proposed Markov random field framework. The algorithm has been incorporated into a live system with robust results for different challenging outdoor pool conditions.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"1 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":"116874150","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.2003.1234053
Yasuaki Nakamura, H. Dekihara
In the engineering database system, multiple versions of a design, including engineering drawings, should be managed efficiently. Spatial data structures can manage spatial objects in a drawing efficiently. The paper proposes extended spatial data structures for efficient management of multiversion engineering drawings. The R-tree is adapted as a basic data structure. The efficient mechanism to manage the difference between drawings is introduced to the R-tree to eliminate redundant duplications and to reduce the amount of storage required for the data structure. Extended data structures of the R-tree, called MVR and MVR* trees, are developed and the performances of these trees are evaluated. A series of simulation tests shows that, compared with the basic R-tree, the amounts of storage required for the MVR and MVR* trees are reduced to 50% and 30%, respectively. The search efficiencies of the R, MVR, and MVR* trees are almost the same.
{"title":"Spatial data structures for version management of engineering drawings in CAD database","authors":"Yasuaki Nakamura, H. Dekihara","doi":"10.1109/ICIAP.2003.1234053","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234053","url":null,"abstract":"In the engineering database system, multiple versions of a design, including engineering drawings, should be managed efficiently. Spatial data structures can manage spatial objects in a drawing efficiently. The paper proposes extended spatial data structures for efficient management of multiversion engineering drawings. The R-tree is adapted as a basic data structure. The efficient mechanism to manage the difference between drawings is introduced to the R-tree to eliminate redundant duplications and to reduce the amount of storage required for the data structure. Extended data structures of the R-tree, called MVR and MVR* trees, are developed and the performances of these trees are evaluated. A series of simulation tests shows that, compared with the basic R-tree, the amounts of storage required for the MVR and MVR* trees are reduced to 50% and 30%, respectively. The search efficiencies of the R, MVR, and MVR* trees are almost the same.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"47 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":"132011744","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.2003.1234115
S. Coleman, B. Scotney, M. G. Herron
The problem of edge evaluation in relation to image gradient-based edge detectors has been widely studied, and there exist a range of edge evaluation techniques that are appropriate to such edge detectors. Although discrete second derivative operators often form the basis of edge detection methods, whereby zero-crossings are used to locate edge pixels, rather less attention has been paid to the development of edge evaluation techniques that are directly appropriate to zero-crossing methods. We propose a new evaluation technique that performs edge sensitivity analysis with respect to angular orientation and displacement errors for edges located by such discrete second derivative operators. The technique applies a finite element interpolation to the output values of the second derivative operator. Hence the method is used to directly evaluate edges located by a second derivative operator without the need to use a supplementary first derivative operator for gradient approximation.
{"title":"An empirical performance evaluation technique for discrete second derivative edge detectors","authors":"S. Coleman, B. Scotney, M. G. Herron","doi":"10.1109/ICIAP.2003.1234115","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234115","url":null,"abstract":"The problem of edge evaluation in relation to image gradient-based edge detectors has been widely studied, and there exist a range of edge evaluation techniques that are appropriate to such edge detectors. Although discrete second derivative operators often form the basis of edge detection methods, whereby zero-crossings are used to locate edge pixels, rather less attention has been paid to the development of edge evaluation techniques that are directly appropriate to zero-crossing methods. We propose a new evaluation technique that performs edge sensitivity analysis with respect to angular orientation and displacement errors for edges located by such discrete second derivative operators. The technique applies a finite element interpolation to the output values of the second derivative operator. Hence the method is used to directly evaluate edges located by a second derivative operator without the need to use a supplementary first derivative operator for gradient approximation.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"10 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":"124496834","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.2003.1234024
M. Bicego, U. Castellani, Vittorio Murino
In this paper, a new system for face recognition is proposed, based on hidden Markov models (HMM) and wavelet coding. A sequence of overlapping sub-images is extracted from each face image, computing the wavelet coefficients for each of them. The whole sequence is then modelled by using hidden Markov models. The proposed method is compared with a DCT coefficient-based approach (Kohir et al. (1998)), showing comparable results. By using an accurate model selection procedure, we show that results proposed in Kohir can be improved even more. The obtained results outperform all results presented in the literature on the Olivetti Research Laboratory (ORL) face database, reaching a 100% recognition rate. This performance proves the suitability of HMM to deal with the new JPEG2000 image compression standard.
本文提出了一种基于隐马尔可夫模型和小波编码的人脸识别系统。从每个人脸图像中提取重叠子图像序列,计算每个子图像的小波系数。然后用隐马尔可夫模型对整个序列进行建模。将所提出的方法与基于DCT系数的方法(Kohir et al.(1998))进行了比较,结果具有可比性。通过使用精确的模型选择程序,我们表明Kohir提出的结果可以得到更大的改进。获得的结果优于Olivetti研究实验室(ORL)人脸数据库上的所有文献结果,达到100%的识别率。这一性能证明了隐马尔可夫算法在JPEG2000图像压缩标准下的适用性。
{"title":"Using hidden Markov models and wavelets for face recognition","authors":"M. Bicego, U. Castellani, Vittorio Murino","doi":"10.1109/ICIAP.2003.1234024","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234024","url":null,"abstract":"In this paper, a new system for face recognition is proposed, based on hidden Markov models (HMM) and wavelet coding. A sequence of overlapping sub-images is extracted from each face image, computing the wavelet coefficients for each of them. The whole sequence is then modelled by using hidden Markov models. The proposed method is compared with a DCT coefficient-based approach (Kohir et al. (1998)), showing comparable results. By using an accurate model selection procedure, we show that results proposed in Kohir can be improved even more. The obtained results outperform all results presented in the literature on the Olivetti Research Laboratory (ORL) face database, reaching a 100% recognition rate. This performance proves the suitability of HMM to deal with the new JPEG2000 image compression standard.","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":"114831234","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.2003.1234021
H. Ragheb, E. Hancock
We illustrate the use of the Beckmann formulation of the Kirchhoff theory for surface analysis problems in computer vision. The Beckmann model is a physical model that describes the reflectance of light from rough surfaces. Here, we use the modified form of the Beckmann model for slightly-rough surfaces using the modification of C.L. Vernold and J.E. Harvey (see Proc. SPIE, vol.3426, p.51-6, 1998). The parameters of the model are the surface roughness and the correlation length. We show how the surface roughness can be estimated using the specular reflectance properties. We also propose a technique for estimating the correlation length using pairs of surface images, subject to different illumination directions. With these parameters to hand, the Beckmann model may be used to perform photometric correction, and hence shape-from-shading may be applied to the corrected Lambertian image to recover improved shape. This model may also be used to re-illuminate the recovered surface. We present experiments to illustrate the utility of the method for each of these tasks.
{"title":"Improving shape recovery by estimating properties of slightly-rough surfaces","authors":"H. Ragheb, E. Hancock","doi":"10.1109/ICIAP.2003.1234021","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234021","url":null,"abstract":"We illustrate the use of the Beckmann formulation of the Kirchhoff theory for surface analysis problems in computer vision. The Beckmann model is a physical model that describes the reflectance of light from rough surfaces. Here, we use the modified form of the Beckmann model for slightly-rough surfaces using the modification of C.L. Vernold and J.E. Harvey (see Proc. SPIE, vol.3426, p.51-6, 1998). The parameters of the model are the surface roughness and the correlation length. We show how the surface roughness can be estimated using the specular reflectance properties. We also propose a technique for estimating the correlation length using pairs of surface images, subject to different illumination directions. With these parameters to hand, the Beckmann model may be used to perform photometric correction, and hence shape-from-shading may be applied to the corrected Lambertian image to recover improved shape. This model may also be used to re-illuminate the recovered surface. We present experiments to illustrate the utility of the method for each of these tasks.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"17 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":"116645598","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.2003.1234073
S. Svensson, C. Arcelli, G. S. D. Baja
Topological properties are global features that can be useful for recognition of digital objects. For example, this is the case for objects having a complex shape without being decomposable into meaningful simple parts. In the case of 3D binary images, topological features are the object components, cavities, and tunnels. While object components and cavities are easy to define and identify, to our knowledge, no computationally convenient way to find tunnels is available. The aim of the paper is to fill this gap by presenting a convenient procedure to detect and represent tunnels in 3D objects.
{"title":"Finding cavities and tunnels in 3D complex objects","authors":"S. Svensson, C. Arcelli, G. S. D. Baja","doi":"10.1109/ICIAP.2003.1234073","DOIUrl":"https://doi.org/10.1109/ICIAP.2003.1234073","url":null,"abstract":"Topological properties are global features that can be useful for recognition of digital objects. For example, this is the case for objects having a complex shape without being decomposable into meaningful simple parts. In the case of 3D binary images, topological features are the object components, cavities, and tunnels. While object components and cavities are easy to define and identify, to our knowledge, no computationally convenient way to find tunnels is available. The aim of the paper is to fill this gap by presenting a convenient procedure to detect and represent tunnels in 3D objects.","PeriodicalId":218076,"journal":{"name":"12th International Conference on Image Analysis and Processing, 2003.Proceedings.","volume":"85 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":"129449656","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}