Image retrieval consists in finding among a database of target images those which represent the same scene as this represented by a query image. In this context, the key-point is the extraction of efficient descriptors from the images. Several recent studies have shown that color histograms can be very efficient in this context. However, color histograms do not take into account the spatial arrangement of colors and are very sensitive to illumination changes.
Several authors have shown that the spatial organization of colors could significantly improve the performance of the image classifiers. Some spatial descriptors, such as color correlograms, may be used to represent in a global way the spatial arrangement of colors in the image, but they are sensitive to object occlusion and scale variation and are not designed to recognize objects in cluttered scenes. In very recent works, few searchers propose to extract local descriptors such as SIFT or SURF which have been proven to be more robust to local features. The SURF outperform the SIFT and are less time consuming. However, the SURF are not yet designed to work on color images.
Objectives:
The aim of this thesis is to extend the SURF to: (a) color information, (b) interest point detection and (c) local description. The interest points detected by the SURF approach result from the analysis of the intensity function discontinuities. However, it has been shown that, in the context of fast object recognition, the human visual system is attracted by regions characterized by high visual saliency and that these visual attention regions are not necessarily regions characterized by intensity function discontinuities. On the other hand, we propose to detect interest points by exploiting the color visual saliency in the images. Previous works done in our lab. will be used in this context.Furthermore, the SURF descriptors are based on specific local intensity gradients. Since color descriptors are more distinctive than gray level ones, we propose to determine new local color descriptors that are invariant to illumination changes. Previous works done in our lab. will be used in this context.
Candidacy:
We seek strong and motivated candidates. Selection will be based on: 1. Excellence of the candidate: outstanding achievement in the applicant’s Master of Science degree level in imaging science, computer science, or any discipline pertaining to the quantitative description of image processing, provided that the applicant can provide evidence of the necessary previous knowledge (i.e. a base of minimal competencies) particularly in the fields of computer science fundamentals, and image analysis and signal processing fundamentals. 2. Language ability: the candidates must demonstrate sound knowledge of the language (the requirement for competence in English is equivalent to TOEFL with at least 213 points (computer based)/550 points (paper based) or IELTS at grade 6.5. Knowledge of French will be not compulsory, but will be beneficial. 3. Student motivation to undertake the PhD and relevance to his/her professional development (explaining the application, the present situation, the interest in the PhD degree, the intentions after this PhD degree, …).
The application deadline is June 1, 2008
To receive the application form and/or to get more information please contact :
Alain TREMEAU, Alain.Tremeau[ at ]univ-st-etienne.fr
Several authors have shown that the spatial organization of colors could significantly improve the performance of the image classifiers. Some spatial descriptors, such as color correlograms, may be used to represent in a global way the spatial arrangement of colors in the image, but they are sensitive to object occlusion and scale variation and are not designed to recognize objects in cluttered scenes. In very recent works, few searchers propose to extract local descriptors such as SIFT or SURF which have been proven to be more robust to local features. The SURF outperform the SIFT and are less time consuming. However, the SURF are not yet designed to work on color images.
Objectives:
The aim of this thesis is to extend the SURF to: (a) color information, (b) interest point detection and (c) local description. The interest points detected by the SURF approach result from the analysis of the intensity function discontinuities. However, it has been shown that, in the context of fast object recognition, the human visual system is attracted by regions characterized by high visual saliency and that these visual attention regions are not necessarily regions characterized by intensity function discontinuities. On the other hand, we propose to detect interest points by exploiting the color visual saliency in the images. Previous works done in our lab. will be used in this context.Furthermore, the SURF descriptors are based on specific local intensity gradients. Since color descriptors are more distinctive than gray level ones, we propose to determine new local color descriptors that are invariant to illumination changes. Previous works done in our lab. will be used in this context.
Candidacy:
We seek strong and motivated candidates. Selection will be based on: 1. Excellence of the candidate: outstanding achievement in the applicant’s Master of Science degree level in imaging science, computer science, or any discipline pertaining to the quantitative description of image processing, provided that the applicant can provide evidence of the necessary previous knowledge (i.e. a base of minimal competencies) particularly in the fields of computer science fundamentals, and image analysis and signal processing fundamentals. 2. Language ability: the candidates must demonstrate sound knowledge of the language (the requirement for competence in English is equivalent to TOEFL with at least 213 points (computer based)/550 points (paper based) or IELTS at grade 6.5. Knowledge of French will be not compulsory, but will be beneficial. 3. Student motivation to undertake the PhD and relevance to his/her professional development (explaining the application, the present situation, the interest in the PhD degree, the intentions after this PhD degree, …).
The application deadline is June 1, 2008
To receive the application form and/or to get more information please contact :
Alain TREMEAU, Alain.Tremeau[ at ]univ-st-etienne.fr
Comments