Composite source modeling based on VQ and arithmetic coding for digital subband video compression

We describe a new source model for entropy coding of subband images. It considers and exploits the nonstationary character and the statistical dependencies present in such data in order to achieve higher compression compared to classical methods based on zero order modeling of the subbands. Indeed, images are made up of various regions with different local power spectra, or textures, separated by edges. Therefore, after subband decomposition, the complex and variable statistical dependencies across the subbands at a particular spatial location should be exploited by a proper modeling in order to achieve lower bit costs. A composite source model is proposed for this purpose. It is based on the combination of vector quantization and entropy-coded scalar quantization. An application to digital video compression is also described. Simulation results show a gain of the composite source model in term of bit rate of around 15% to 20% compared to the classical runlength/Huffman coding of the subband samples