VLBV98 Panel 3:

VLBV 98: Oct 8-9, 1998, University of Illinois at Champaign/Urbana


Robert Forchheimer

 When first conceived in the early 80s, Model-Based Image Coding was a technique which had very little in common with the virtual reality concepts that started to spread rapidly at the latter part of that decade. Rather, the idea behind model-based coding was to reduce the amount of information required to transmit the image of a talking person. The principal problems discussed at that time dealt with how to automatically find out the shape and motion of the human head from a camera that generated a video sequence. The information that needed to be sent was expected to be in the low kbit/s range.

Contrary to this view, the early efforts in VR was to render 3D scenes of high complexity such that they would look as realistic as possible. Typically there was no "ground truth" which had to be obeyed except that physical laws should be considered when objects were moving or touching each other. Data were expected to be carried by wide-band local area networks and counted in the Mbit/s range.

Despite the large differences between the two concepts, they are both based on the same underlying principle, namely computer modeling of 3D objects. This common denominator became clear in the early 90:s leading to suggestions and research projects to build tele conferencing systems based on the combined techniques of VR and Model-based coding. Also, the close relation between the SNHC group and the Face-and-Body animation groups of MPEG-4 manifested such a common view.

However, as pointed out above, there are also major differences between the two techniques. These differences became quite clear during the MPEG-4 work. It was obvious that goals, critical technical issues and implementation philosophies differed considerably between the two techniques.

Panel 3 of VLBV'98 will discuss "3D modeling and VR" in the context of Very Low Bitrate Video Coding. This means that the above issues may well surface and that they may even be sharpened as model-based coding is extended to handle more general objects than just the human face. It is thus my hope that the presentations of the panelists together with the following discussion will shead more light on the above mentioned differences and similarities.