Face Mosaicing for Pose Robust Video-Based Recognition

This paper proposes a novel face mosaicing approach to modeling human facial appearance and geometry in a unified framework. The human head geometry is approximated with a 3D ellipsoid model. Multi-view face images are back projected onto the surface of the ellipsoid, and the surface texture map is decomposed into an array of local patches, which are allowed to move locally in order to achieve better correspondences among multiple views. Finally the corresponding patches are trained to model facial appearance. And a deviation model obtained from patch movements is used to model the face geometry. Our approach is applied to pose robust face recognition. Using the CMU PIE database, we show experimentally that the proposed algorithm provides better performance than the baseline algorithms. We also extend our approach to video-based face recognition and test it on the Face In Action database.