Statistical coding method for facial features

Model-based techniques have been shown to give high compression rates for coding head and shoulder image sequences, typically for videophone applications. However, they lead to poor image quality in significant areas of the face such as the eyes and mouth. To overcome this problem, a hybrid system could be perceived where the facial features were represented using traditional statistical techniques and the remaining of the head and shoulder sequences using highly efficient model-based methods, therefore utilising more bits to code the sensitive areas and fewer for the rest. In the paper, the method of principal component analysis to code the dynamic changes in a sequence is presented