Lossless Video Coding Using Lattice Based Distributed Source Coding Techniques

Information theoretic proof exists to support that independent encoding of distributed sources with a joint decoder by exploiting the correlation among the sources can be as efficient as encoding them jointly. This paper successfully attempts to apply this concept for lossless video coding using lattices to divide the multidimensional integer pixel intensity hyperspace of a block of pixels into a finite number of cosets and encoding each block with its coset index. This radical departure from conventional predictive coding techniques not only offers very low computational complexity by avoiding expensive learning of predictor coefficients but also avoids any coding loss due to rounding of real-valued predictions. On standard test video sequences, this scheme achieved compression within as low as 4.6% of the latest scheme with optimal learning of predictor coefficients. The former, however, outperformed the latter as soon as it started updating the optimal predictor coefficients less frequently to reduce the computational complexity.