Three dimensional subband video analysis and synthesis with adaptive clustering in high frequency subbands

This paper presents a hybrid analysis and synthesis scheme for video coding aimed at developing a video communication system that can consistently transmit high quality video signals with high compression ratios. The scheme is based on a 3D subband decomposition of video signal. To best take advantage of the characteristics of high frequency subbands, an adaptive clustering algorithm with Gibbs random field model is applied to the high frequency subbands to achieve an adaptive quantization with spatial constraints. The adaptive quantization reduces significantly the activity of high frequency bands and makes the entire scheme amenable to efficient video coding. The resultant reconstruction noise can be suppressed using again a Gibbs random field based algorithm. Preliminary results have shown the potential of this methods over existing methods.