Pré-analyse de la vidéo pour un codage adapté : application au codage TVHD en flux H.264
Auteur / Autrice : | Olivier Brouard |
Direction : | Dominique Barba, Vincent Ricordel |
Type : | Thèse de doctorat |
Discipline(s) : | Automatique, robotique, traitement du signal et informatique appliquée |
Date : | Soutenance en 2010 |
Etablissement(s) : | Nantes |
Ecole(s) doctorale(s) : | École doctorale sciences et technologies de l'information et des matériaux (Nantes) |
Partenaire(s) de recherche : | autre partenaire : École polytechnique de l'Université de Nantes |
Mots clés
Mots clés contrôlés
Mots clés libres
Résumé
The optimization methods of a classical video encoder process the information to reduce it only as a signal point of view, and are therefore «low level». Although some works incorporating the HVS properties have been proposed to assess the quality, or to improve the coding, these methods have not been included in the standards. Researchs focus more on the improvment of new standards, such as H. 264. However, the performances of the «high level» methods are encouraging. We therefore propose a method of pre-analysis of videos, that incorporates a visual attention model. This model aims at analyzing the video taking into account the high level information, in order to transmit to the encoder an optimal set of parameters, and to exploit efficiently the coding tools. Researchs have highlighted the fundamental characteristic of the motion contrast for the visual attention. Our pre-analysis method first detects the moving objects (relative to that of the camera), then calculates the saliency maps to localize the visually important areas. We propose two coding applications (that can be exploited jointly) based on information obtained after the pre-analysis, and the assessment of their performance. The first proposes to adaptively modify the structure of the GOP, it is based on the spatio-temporal content of the video. The second concerns an application of video compression with a visual differentiated quality guided by the saliency maps. The performances of these two methods are carried out using tests of subjective quality assessment