Facial features in image sequences of a single subject are coded using principal components. Since the class of images used is restricted, a very efficient coding can be obtained and mouth images have been coded with reasonable quality at intraframe compression ratios of about 400.