Search

International audience; Holographic data plays a crucial role in the recent three dimensional imaging as well as microscopic applications. As a result, huge amounts of storage capacity will be involved for this kind of data. Therefore, it becomes necessary to develop efficient hologram compression schemes for storage and transmission purposes. In this paper, we focus on the shifted distance information, obtained by the phase shifting algorithm, where two sets of difference data need to be encoded. More precisely, a non separable vector lifting scheme is investigated in order to exploit the two-dimensional characteristics of the holographic contents. Simulations performed on different digital holograms have shown the effectiveness of the proposed method in terms of bitrate saving and quality of objectreconstruction.

International audience; Many existing works related to lossy-to-lossless image compression are based on the lifting concept. In this paper, we present a sparse op- timization technique based on recent convex algorithms and applied to the prediction filters of a two-dimensional non separable lifting structure. The idea consists of designing these filters, at each resolution level, by minimizing the sum of the ℓ1-norm of the three detail subbands. Extending this optimization method in order to perform a global minimization over all resolution levels leads to a new opti- mization criterion taking into account linear dependencies between the generated coefficients. Simulations carried out on still images show the benefits which can be drawn from the proposed optimization techniques.

International audience; Due to the great interest of stereo images in several applications, it becomes mandatory to improve the efficiency of existing coding techniques. For this purpose, many research works have been developed. The basic idea behind most of the reported methods consists of applying an inter-view prediction via the estimated disparity map followed by a separable wavelet transform. In this paper, we propose to use a two-dimensional non separable decomposition based on the concept of vector lifting scheme. Furthermore, we focus on the optimization of all the lifting operators employed with the left and right images. Experimental results carried out on different stereo images show the benefits which can be drawn from the proposed coding method.

National audience; Dans cet article, nous nous intéressons à l'optimisation des filtres de prédiction et de mise à jour apparaissant dans un schéma de lifting 2D non-séparable par des méthodes d'optimisation convexe non lisse. Comparativement à l'approche usuelle reposant sur la minimisation de la norme euclidienne L-2, la méthode proposée permet de minimiser des critères parcimonieux. Plus précisément, nous proposons de nouveaux algorithmes efficaces pour minimiser des critères de type L-1 et L-1 pondéré.

This thesis addresses the problem of stereo image coding for storage and transmission purposes. In particular, we are interested in lossy-to-lossless coding schemes allowing progressive reconstruction. In the first part, we propose novel coding methods based on Vector Lifting Schemes. Unlike conventional approaches which encode a reference image, a residual one and a disparity map, the proposed methods do not generate a residual image but two compact multiresolution representations of the left and the right images simultaneously. We also show the benefits that can be drawn from integrating a smooth and dense disparity field within such joint stereo image coding schemes. In the second part, we propose optimization techniques which can be used in the design of these lifting schemes. This allows us to build content adaptive methods. More precisely, we focus on the optimization of all the operators (i.e the update and prediction filters) involved in a lifting structure. To this end, we propose and analyze a new criterion for the design of the update filter. Concerning the prediction filters, we investigate techniques for optimizing sparsity criteria., L'objectif majeur de ce travail de thèse était de proposer et d'analyser de nouveaux schémas de codage d'images stéréoscopiques. Ces schémas permettent d'assurer la reconstruction progressive avec la possibilité de restitution exacte de la paire d'images. Dans une première partie, nous avons proposé de nouveaux schémas de codage conjoint de la paire d'images stéréo reposant sur le concept du lifting vectoriel. En ce sens, contrairement aux méthodes classiques, la méthode proposée ne génère aucune image résiduelle mais deux représentations multirésolutions compactes de l'image gauche et l'image droite. De plus, nous avons proposé d'intégrer une méthode récente d'estimation de disparité dans des applications de codage d'images stéréoscopiques. Puis, dans une deuxième partie, nous nous sommes intéressés aux schémas de lifting 2D non séparables tout en nous focalisant sur les aspects d'optimisation des différents filtres mis en jeu. La structure de lifting 2D considérée est composée de trois étapes de prédiction suivies par une étape de mise à jour. Plus précisément, nous avons proposé et analysé une nouvelle méthode pour la conception du filtre de mise à jour. Concernant les filtres de prédiction, nous avons développé de nouveaux critères parcimonieux de types L1.

Publication in the conference proceedings of EUSIPCO, Lisbon, Portugal, 2014

International audience; In this paper, we are interested in studying the impact of the compression on the performances of some wavelet-based retrieval systems. Firstly, we show that the quantization operation of JPEG2000 has a negative effect on the performance of these contentbased image retrieval systems for a given feature extraction method. Moreover, in this work, we aim at designing a novel retrieval strategy in order to reduce the performance drop resulting from quantizing the query and database images at different bitrates. More precisely, instead of directly comparing these images, we propose to perform the comparison step at similar qualities by re-compressing the high quality image at the bitrate of the low quality one. Experimental results corroborate the gain achieved by the proposed strategy.

Publication in the conference proceedings of EUSIPCO, Bucharest, Romania, 2012

International audience; Many existing works related to lossy-to-lossless image compressionare based on the lifting concept. In this paper, we present a sparse optimization technique based on recent convex algorithms and appliedto the prediction filters of a two-dimensional non separable liftingstructure. The idea consists of designing these filters, at each resolution level, by minimizing the sum of the ℓ1-norm of the three detailsubbands. Extending this optimization method in order to performa global minimization over all resolution levels leads to a new optimization criterion taking into account linear dependencies betweenthe generated coefficients. Simulations carried out on still imagesshow the benefits which can be drawn from the proposed optimization techniques.

This thesis addresses the problem of stereo image coding for storage and transmission purposes. In particular, we are interested in lossy-to-lossless coding schemes allowing progressive reconstruction. In the first part, we propose novel coding methods based on Vector Lifting Schemes. Unlike conventional approaches which encode a reference image, a residual one and a disparity map, the proposed methods do not generate a residual image but two compact multiresolution representations of the left and the right images simultaneously. We also show the benefits that can be drawn from integrating a smooth and dense disparity field within such joint stereo image coding schemes. In the second part, we propose optimization techniques which can be used in the design of these lifting schemes. This allows us to build content adaptive methods. More precisely, we focus on the optimization of all the operators (i.e the update and prediction filters) involved in a lifting structure. To this end, we propose and analyze a new criterion for the design of the update filter. Concerning the prediction filters, we investigate techniques for optimizing sparsity criteria.; L'objectif majeur de ce travail de thèse était de proposer et d'analyser de nouveaux schémas de codage d'images stéréoscopiques. Ces schémas permettent d'assurer la reconstruction progressive avec la possibilité de restitution exacte de la paire d'images. Dans une première partie, nous avons proposé de nouveaux schémas de codage conjoint de la paire d'images stéréo reposant sur le concept du lifting vectoriel. En ce sens, contrairement aux méthodes classiques, la méthode proposée ne génère aucune image résiduelle mais deux représentations multirésolutions compactes de l'image gauche et l'image droite. De plus, nous avons proposé d'intégrer une méthode récente d'estimation de disparité dans des applications de codage d'images stéréoscopiques. Puis, dans une deuxième partie, nous nous sommes intéressés aux schémas de lifting 2D non séparables tout en nous focalisant sur les aspects d'optimisation des différents filtres mis en jeu. La structure de lifting 2D considérée est composée de trois étapes de prédiction suivies par une étape de mise à jour. Plus précisément, nous avons proposé et analysé une nouvelle méthode pour la conception du filtre de mise à jour. Concernant les filtres de prédiction, nous avons développé de nouveaux critères parcimonieux de types L1.