Principal component coding of mouth sequences

D. Shah, S. Marshall, W. J. Welsh
1993 IEEE International Conference on Acoustics, Speech, and Signal Processing, Minneapolis, MN, USA, 1993, Volume 5, pp. 373-376
Facial features of a videophone image sequence are coded using the traditional technique of principal components (PCs). One of the main advantages of this method is that it makes use of a large a priori knowledge of the images and exploits the correlation between pixel values. The sequences are coded through a hybrid wire frame model in which the head and shoulder images are coded using a small number of bits. A box is placed around the mouth and eye regions which are coded at a higher bit rate using PCs. Since the image sequence is similar in its overall configuration, this scheme gives a very high compression ratio of around 500:1.