In audio visual speech recognition and lipreading, the widely used ASM (Active Shape Model) for lip contour extraction suffers from the lack of robustness and cannot extract the exact lip contours due to the various mouth shape changes when uttering.