Face recognition from video: a CONDENSATION approach

The aim of this work is to investigate how to exploit the temporal information in a video sequence for the task of face recognition. Following the approach in (Li and Chellappa, 2000), we propose a probabilistic model parameterized by a tracking state vector and a recognizing identity variable, simultaneously characterizing the kinematics and identity of humans. We then invoke a CONDENSATION (Isard and Blake, 1996) approach to provide a numerical solution to the model. Once the joint posterior distribution of the state vector and the identity variable is estimated, we marginalize it over the state vector to yield a robust estimate of the posterior distribution of the identity variable. Due to the propagation of identity and dynamics, a degeneracy in the posterior distribution of the identity variable is achieved to give improved recognition. This evolving behavior is characterized using changes in entropy. The effectiveness of this approach is illustrated using experimental results on low-resolution video data.