Techniques de codage d'images basées représentations parcimonieuses de scènes et prédiction spatiale multi-patches

摘要

In recent years, video compression eld has increased signicantly since the apparitionof H.264/AVC standard and of its successor HEVC. Spatial prediction in these standardsare based on the unidirectional propagation of neighboring pixels. Although very effectiveto extend pattern with the same characteristics, this prediction has limited performances toextrapolate complex textures. This thesis aims at exploring new spatial prediction schemesto improve the current intra prediction techniques, by extending these local schemes toglobal, multidimensional and multi-patches schemes. A hybrid prediction method based ontemplate and block matching is first investigated. This hybrid approach is then extended tomulti-patchs-based prediction of type "Neighbor Embedding" (NE). The other part of thisthesis is dedicated to the study of epitome image within the scope of image compression.The idea is to exploit spatial redundancies in the original image in order to rst extracta summary image containing the texture patches the most representative of the image,and then use this compacted representation to rebuild the original image. The conceptof epitome has been incorporated in two compression schemes, one of these algorithms isin rupture with the traditional techniques since the image blocks are processed, both atencoder and decoder sides, in a spatial order that depends on the image content and this inthe interest of propagating image structures. In this last compression algorithm, extendedH.264 Intra directional prediction modes and advanced multi-patches prediction methodshave been also included. These different solutions have been integrated in a H.264/AVCencoder in order to assess their coding performances with respect to H.264 intra modesand the state of the art relative to these dierent techniques.