Evaluation des mouvements rigides et non rigides du visage pour les videoconferences (French text)

The goal of videoconference is to allow two or many users to communicate far away from each other by using computer to have sounds and images. Since there is a lot of data in an image sequence and redundancy in the signal, many techniques have been studied to reduce the transmission rate. When the images are concerned in general scenes, the techniques often consist to compress data, without using any real scene model. In the case of a "head-and-shoulders" videoconference, a lot of information from the scene is available because that's a face user and many anthropometric constraints are nearly constant for most of the population. So the goal of this project is to define an images sequences specialized system to read the human head and face motions in an entirely automatic way. (Abstract shortened by UMI.)...