De la généricité à la sélectivité des descripteurs vidéo : application à la détection de copies par le contenu
Auteur / Autrice : | Julien Law-To |
Direction : | Nozha Ben Hajel-Boujemaa |
Type : | Thèse de doctorat |
Discipline(s) : | Informatique |
Date : | Soutenance en 2007 |
Etablissement(s) : | Versailles-St Quentin en Yvelines |
Mots clés
Mots clés contrôlés
Résumé
My PhD thesis presents a new approach for indexing large sets of videos by their content. The proposed concept is based on the extraction and the local description of different natures of points of interest and further on the estimation of their trajectories along the video sequence. Analyzing the low-level description obtained allows highlighting semantic labels of behaviors. Searching for copies in large video databases is a new critical issue. ViCopT is a system dedicated to video copy detection based on our video description. A complete evaluation on a large video database (1,000 hours) demonstrates the robustness and the discriminability of ViCopT and the relevance of our strategy. Comparative evaluations in European and international contexts present the high performances of our system facing other academic and industrial systems.