Trajectory-Based Video Indexing and Retrieval Enabling Relevance Feedback

LES AUTEURS : Le Thi Lan, Alain Boucher, Monique Thonnat

RESUME:

This paper is proposing an approach for retrieving videos based on object trajectories. First, a trajectory is translated into a sequence of symbols based on a symbolic representation, beyond the initial numeric representation, which does not suffer from scaling, translation or rotation. Then, in order to compare trajectories based on their symbolic representations, two similarity measures are proposed, inspired by works in bioinformatic. Moreover, based on these similarity measures, two relevance feedback strategies are given. Experimental results for two databases show that the proposed similarity measures gave results as good as other existing measures. Real advantages of these measures are the possibility for the partial matching and for relevance feedback.

Mots clé: Video Indexing and Retrieval, Trajectory Matching, Relevance Feedback

Pour télécharger cet article, cliquez ici

BibTeX reference:

@INPROCEEDINGS{lle_ICCE06,
author = {Le Thi Lan, Alain Boucher, Monique Thonnat},
title = {Trajectory-Based Video Indexing and Retrieval Enabling
Relevance Feedback},
booktitle = {First International Conference on Communications and
Electronics (ICCE'06)},
year = {2006},
address = {Hanoi, Vietnam},
month = {October},
}

Dernière mise à jour : 23/11/07
Catherine.Martin@sophia.inria.fr