Efficient context modeling in scalable 3D wavelet-based video compression

We address the problem of context modeling and coding in the framework of a three dimensional wavelet-based video compression scheme with temporal filtering in the motion direction. Embedded coding algorithms based on a 3D decomposition (like the 3D SPIHT algorithm) provide scalability, which is an important feature for delivering compressed bitstreams over heterogeneous networks. We propose a highly scalable video compression scheme, based on an efficient contextual modeling and a hierarchical encoding of the spatio-temporal trees of wavelet coefficients. We have shown that the efficiency of the final arithmetic encoding stage is increased by performing a statistical modeling based on the context tree weighting method. In order to reflect changes in the statistical behaviors of the different information sources that contribute to the bitstream, several statistical models have been distinguished. As shown by simulations, the proposed method results in significant improvements of the coding efficiency, while keeping the scalability functionality