Codage video distribué

Abstract

An important property of Wyner-Ziv video coders is the shift of motion estimation to the decoder. This operation is performed on intra key frames and interpolated to Wyne-Ziv coded frames. In this thesis, we propose a new block-based architecture of a Wyner-Ziv coder that does not use key frames and exploits partial matching motion estimation at the decoder. A small part of each macroblock is intra coded first with a given quality and the other parts are gradually coded. The decoder can thus try to estimate the real motion of the macroblock and gradually improved motion is used to generate better side information. In addition to being more practical, the simulations with the proposed coder show promising results and good opportunities to bridge the efficiency gape between the Wyner-Ziv coders and conventional inter coding and also a practical robust Wyner-Ziv coder to channel losses