A Robust Real Time Face Location Algorithm for Videophones

When videophone images are compressed to low bit rates (< 64 kbits/s) for transmission over telephone lines, the image quality deteriorates when there is a great deal of movement in the frame. If the coder has knowledge of the videophone user's location, this information can be used to enhance the area of interest (the user) at the expense of the unchanging and/or uninteresting background. It has been shown in subjective tests that such enhancement greatly improves the picture quality at such bit rates. The author describes a computationally inexpensive algorithm to enable videophones to track the user for the purpose of such enhancement.