Initialization of Model Based Coding in Adaptive Video Transmission:An Analysis-By-Synthesis approach

This paper consider the problem of video content adaptation for the teleconference application in a heterogeneous network communication system. Due to the diversity of content providers and content consumers in the network, the content of the network transportation is often preferred to be adaptive for both efficiency and reliability in the communication. The video content adaptation could be achieved by using normal video codec when network transmission is smooth and switching to a very low bit rate Model based coding (MBC) codec when network congestion happen. An inevitable step of this switching is the initialization of the MBC, that is, to fit a genetic face model onto the first video frame. This paper revised the Analysis-by-synthesis(ABS) approach used in many vision based tracking problems and report our strategy of implementing the initialization using an ABS approach. A perturbation based global optimization algorithm, Simulated annealing (SA) is examined within the ABS framework. Problems are identified and different ways are studied, promising performance results of our suggested approach are reported.