Facial feature tracking combining model-based and model-free method

In this paper, we propose a novel facial feature tracker, which integrates model-free and model-based method to reliably track dense facial features with complex non-rigid motions in low quality video sequence. The tracker consists of the mouth motion model part and the enhanced KLT tracker part. There are three key elements in our algorithms: dense facial feature tracking, noise removal with global rank constraints and characteristic non-rigid motion description. Experiments show good results on tracking dense facial features under various expressions, even some facial features have degenerate features.