Adaptive Gaussian Mixture Models Based Facial Actions Tracking

Recently adaptive Gaussian mixture models have become increasingly popular on account of their strong ability to adapt to variations. In this paper, an algorithm based on adaptive mixture models is proposed to track facial actions in video. WSF Mixture Appearance Model is taken to depict image observation and an active learning scheme which combines fast convergence and temporal adaptability is presented. A 3d parameterized model is used to model the face and facial actions, mixture observation model is built on shape free texture, and then a gradient descend fitting algorithm is taken to track parameters. Experiments demonstrate that the algorithm is robust and efficient.