Statistical Learning and Analysis for Unconstrained Face Recognition

Опубликовано 6 сентября 2016, 5:04

Although face recognition has been actively studied during the nineties, the state-of-the-art recognition systems perform poorly when confronted with unconstrained scenarios such as illumination and pose variations, surveillance video, etc. In this talk, we address these challenges by introducing approaches to recognizing human faces under illumination and pose variations and from video sequences, using statistical learning and analysis techniques. For face recognition across illumination, we present a generalized photometric stereo approach by modeling all face appearances belonging to all humans under all lighting conditions. Using a linear generalization, we achieve a factorization of the observation matrix consisting of face appearances of different individuals, each under a different illumination. We resolve ambiguities in factorization using surface integrability and face symmetry constraints. In addition, an illumination-invariant identity descriptor is provided to perform face recognition across illuminations. We further extend the generalized photometric stereo approach to an illuminating light field approach, which is able to recognize faces under pose and illumination variations. For video-based face recognition, while conventional approaches treat tracking and recognition separately, we present a simultaneous tracking-and-recognition approach. This simultaneous approach solved using the sequential importance sampling algorithm, i.e. the particle filter, improves accuracy in both tracking and recognition. In addition, we present a generic tracking algorithm that models online appearance changes in a video sequence using a mixture appearance model and produces good tracking results on various challenging scenarios.