Which problems are you working on now?

We are currently examining the problems of video teleconferencing and video phones, where it is assumed that the most interesting object to send and receive is the human face.

What methods are you using to solve the problems?

We are examining two methods: (1) Minimal 3D models and (2) Vector quantization. In both methods, we are working with triangulation/polygon representations because these are often implemented in off-the-shelf hardware. In the first method, we are trying to determine the effectiveness of different algorithms for minimizing the number of polygons for representing a 3D model. By minimizing the number of polygons, we expect to reduce the bitrate for a given accuracy. In the second method, we are addressing the video phone application in that we assume the camera is fixed and that the primary moving object is the speaker. Thus, the background only needs to be infrequently updated, and most of the bitrate can be utilized for the speaker.

What is the current status of the research? What exists?

Currently, we have developed software for manipulating 3D texture mapped objects, implemented algorithms for minimizing the number of polygons for general 3D models, and have performed tests on real data with respect to minimizing 3D error.