Feature and model tracking for model-based coding

One of the most interesting approaches to communicating image sequences is model-based coding. This is totally unlike conventional image compression and coding, since, rather than transmitting the pixels that comprise an image, it involves the communication of information regarding the motion of components in the scene. The approach to model-based coding discussed here is to employ a mathematical model-usually a 3-D geometric model of the objects in the scene-and to use techniques akin to computer vision to identify the model movement that gave rise to the imagery. Information regarding the model movement is then transmitted in some coded form and image synthesis techniques are used to re-generate the individual frames of the image sequence. Since the volume of data required to describe the motion of a model is normally vastly less than that required to describe an image, model-based techniques could potentially reduce transmission requirements to as little as a few tens of bytes per frame