Compression adaptative de surfaces par ondelettes géométriques

The recent advances in computer graphics and digitization allow access to an ever finer three-dimensional modelling of the world. The critical challenges with 3D models lie in their transmission and rendering, which must fit the heterogeneity of the end resources (network bandwidth, display terminals . . . ). In this context, this thesis investigates the progressive compression and transmission of 3D models, based on multiresolution analysis, to provide a scalable representation of these geometric models. This work is part of "CoSurf", a collaborative research project involving LIRIS laboratory and France Télécom R&D in Rennes. The proposed hierarchical compression method is based on a wavelet decomposition, which is a robust and competitive analysis tool incorporated in the JPEG2000 and MPEG4 standards. The innovation, compared to existing techniques, lies in a prior surface segmentation in regions, each having a different frequency amplitude. Hence, it is possible to treat independently each resulting partition during the wavelet decomposition, quantization and binary allocation. The advantage is to fit the local features of the mesh surface, which can reduce the overall encoding information. It is finally important to consider each patch visual contribution to the entire surface, to optimize the bit allocation, essential for view-dependent transmission and visualization. Other applications like adaptive watermarking, filtering or denoising, indexing or error-resilient coding may benefit from this generic concept we proposedLYON1-BU.Sciences (692662101) / SudocSudocFranceF