Face and 2D Matching
Stephane Marchand-Maillet

Project description

The objective of this project is to design new techniques for the matching of 2D objets within images and video. In particular, we investigate the use of Hidden Markov Models in two dimensions. While real 2D HMMs are too complex to be usable, we introduce pseudo-2D HMMs which provide feasible models. We have defined the basic algorithms for training, and decoding with these models. We apply these models to the problem of face recognition within video sequences. In particular, these models provide a clear approach to face detection and localisation.

We also investigate the use of such models for other applications within video indexing.