An improved transform coder for image sequences using attributes of difference pictures

One of the characteristics of the interframe difference pictures for video teleconferencing types of image sequences is large zero areas and impulse-like non-zero regions. It is possible to use this property to improve ordinary transform coders. In this paper, the method of projection onto convex sets is used to obtain an improved transform coder for impulse-like pictures in general and difference pictures in particular. Although the amount of improvement depends on many factors, including the percentage of motion, it is found significant for many cases.