Représentation en objets vidéo pour un codage progressif et concurrentiel des séquences d'images

The goal of this thesis is to valid the hypothesis such that the video object coding allowssignificant improvement thanks to the used of dynamic coding. The dynamic coding is thesetting on competition of a set of coders on each video object and the retaining of the bestcoder for each object.To reach that goal, different points are studied. The first one concerns the study ofautomatic video object segmentation. We are proposing an object model with the notionof long term tracking. This model represents an object thanks to its motion (by the usedof an active mesh) and its texture (by the used of a mosaic). A 3D clustering algorithm isdeveloped based on that model.In a second part, we are improving object coding techniques via the scalability on thevideo bitstream. To that purpose, we are using a wavelet 3D coding scheme and we arenotably introducing a lossy contour coding.Finally the last point studied concerns the dynamic coding. Coders used are H264/AVC,a 3D wavelet coder, a 3D coder and a mosaic coder. The automatic rate distribution allowsto obtained better results than those given by each coder took separately, in all offering acut-out of the bitstream in video objects.