Content-Based Hybrid DPCM/Classified Vector Quantization for Coding Video Telephony Sequences

Abstract

A novel hybrid DPCM/CVQ (classified vector quantization) method is proposed to encode and decode video telephony sequences. The CVQ coding method is based on the detection of human faces within a video signal. This detection will improve three aspects of video coding efficiency. First, knowledge of the image content can be used to remove coding redundancy within the prediction error. Second, selective coding can be adopted to further reduce psycho-visual redundancy in a video signal. Finally, coding redundancy can be further removed with knowledge of the location of face regions. This coding redundancy reduction is achieved by adopting a 3D run-length-coding method to encode the indexes of the codewords in CVQ. Our bit-rate requirement is 11.41% less than that of the H.261 standard. Furthermore, the quality of our decoded images is about 2 dbs better than the image quality decoded by using the H.261 standard. By sacrificing only very little image quality, our proposed coding method can achieve the bit-rate requirement in very low bit-rate video coding.