Over the last twenty years, much progress has been made in still image based recognition, accompanied by meticulous performance evaluation of many state-of-the art algorithms. It is widely believed that the face recognition problem has been solved for frontal images acquired in controlled illumination conditions. However, when variations due to pose, illumination and aging are present, the performance of many existing algorithms is not good enough for deployment. In this talk, I will discuss three new algorithms for pose and illumination invariant face recognition using still images. These algorithms are derived using generalized photometric stereo, albedo estimation using a non-stationary Wiener filter and pose encoded spherical harmonics. I will then discuss model-based approaches for face recognition across aging in children and adults. Finally, I will discuss the video-based face recognition problem and present two algorithms, one based on the particle filter and the other based on statistical inference on manifolds. The talk will conclude with suggestions for future research directions in this area.
Wednesday, February 06, 2008
Free and open to the public