Video Facial Feature Tracking with Enhanced ASM and Predicted Meanshift

The active shape model (ASM) has been widely used to recognize and track a face from a video sequence. However, it is usually limited to frontal view or the cases of small-scale head movement, as its accuracy may greatly degrade in conditions of quick movement, large rotation and temporary occlusion. We propose an enhanced ASM and predicted mean shift algorithm to meet these challenges, which combines the context information and predicted mean shift to obtain multi-angle start shapes for ASM searching and the best result shape is chosen based on a matching evaluation. Extensive experiments demonstrate the flexibility and accuracy of the proposed method.