Construction de Mosaïques de Super-résolution À Partir de la Vidéo de Basse Résolution: Application Au Résumé Vidéo Et la Dissimulation D'erreurs de Transmission

In this PhD, we are interested in the construction of mosaics from MPEG-1/2 compressed video for the purpose of video summarisation. We directly extract low-resolution images and motion vectors from the compressed stream without complete decoding for the mosaic construction and we apply a super-resolution method in order to increase their resolution and improve their visual quality. Furthermore, we show how the developed methods can be used for other indexing tasks, namely shot boundary detection and camera motion characterisation, and error concealment.