Model-based facial image coding

Abstract

In this thesis, we proposed an automatic system of extracting and tracking non-rigid facial features for the model-based face image coding. First face region is separated from the background using skin color and motion information. Facial features such as eyes and mouth are then located for setting up the initial position for B-snake. We introduce the full parametric motion estimation for the rigid head motion and the B-snake for extraction of non-rigid shape of the facial features. The energy of the B-snake is minimized by applying the greedy algorithm. The movement of the facial features is tracked using the prototype-based deformable template. Multiple image cues such as gradient, region consistency and inter-frame motion are used to facilitate the tracking process. This system is able to capture the subtle changes in facial features for the application of facial expression analysis.