A Monocular Video-Based Facial Expression Recognition System by Combining Static and Dynamic Knowledge

In this paper, a real-time monocular video-based facial expression recognition system is proposed. An adaptively weighted online statistical appearance model is first constructed to retrieve the continuous facial motion information from each frame of input video. The static and dynamic knowledge of facial expression are obtained by facial expression anatomy and manifold learning respectively, and then fused for recognition in framework of particle filtering. The experimental results confirm the advantage of the system, even in the presence of significant head pose and occlusions.