Computer Vision in Next Generation Image and Video Coding

In this paper we will explore the various ways in which Computer Vision techniques can aid and sometimes even replace the classical algorithms used in many still image and video coding standards. Because compression factor is one of the most important performance criteria of image/video coding we will examine the effectiveness of these techniques in achieving low and very-low bit rates. Motivation for the recent demand of very low-bit rate algorithms stems in part from the introduction of advanced multimedia applications. The desktop computer, for instance, is fast becoming much more than a number crunching machine and will likely become the main channel for interpersonal communications as well as our interface to what is being called the information superhighway. The video inputs to such systems will not only need to capture and code images and video but also obtain some measure of understanding of the scenes before them. Classical or first generation coding algorithms, while indispensable in many fields of communications, fall short of this task. Only a fusion of such algorithms with the work currently being done in computer vision, image understanding, and artificial intelligence will produce a viable system. Since, traditionally, these communities have been rather disparate, it is necessary to begin at a basic level and explore the ways in which techniques from each field can be integrated to create a system that is truly more than the sum of its parts. With that end in mind, this paper will address some of the common aspects of image and video coding with a concentration on one of the more important research areas — the estimation of 2D and 3D motion to use in efficiently compressing video sequences using 3D models.