Abstract EN:

Face analysis fields are widely spread out over a large quantity of domains :Human Computer Interaction, security, movies post-production, games. . . It in-cludes detection, recognition, 3D reconstruction, animation, and emotion analy-sis. Face animation was the main motivation of this thesis, and we always kept itin mind at anytime. We discuss here, most of the fields. We first talk about face reconstructionand face modeling with a new model inspired by the Candide Model. Then, weaddress face detection and particularly features detection introducing anthropome-tric constraints in a global scheme through a Markov Random Field formulation. From prior constraints, we are able to estimate the 3d pose of a face from a singleimage and to extend it to motion tracking. We conlude our work with emotion ana-lysis. We propose an expression modelling technique defined as time series, andwe present our 3D database for face expression. We present the state of the art ofemotion recognition. And we finally invoke the use of our expression modellingtechnique as a prediction to be compared with the data in time.