Face Model-based Coding Of Video Communications

The technique of low bit rate image compression has become an active research field because of the rapid development in multimedia, communication and computer and the increase of people's requirement to video serve. The model-based image coding as an analysis-synthesis coding method with high compression and natural reconstructed image effect focuses researchers' eyes, and its theory is represented in the MPEG-4 standard. To the video applications such as videophone meeting or personal mobile communtion in which the main active objects are human faces, the core techniques in model-based image coding such as face recognition in complex scene, facial feature contour estimation, tracking of whole face motion, automatic adaptation of generic face model,and paremeters compression, image synthesis and etc. are deeply studied in this thesis, some results are obtained as following:A fast and effective algorithm to locate face region in video scene whose main active objects are head-shoulders is given.A model of grads area projection curve is built to verify face candidate.The simplified energy functions of eye and mouth deformable templates makes the templates mattching fast and the initial parameters extracted for the minimization process prevent templates from shrinking.A enhenced deformable templte of nose shape is proposed to extract nose contour in face with variant poses fast and exactly.An algorithm to track the chin's contour by combining deformable template with snakes is proposed and corresponding tracking result is stable and accurate.The extracted key points by given corner detection method makes lesser error for face whole motion estimationAn applied scheme is proposed for automatic adaptation of generic face model makes the matching process is fast.The proposed encoding algorithm based on feature region for active texture greatly increases the compression ratio.