Robust distributed video compression based on multilevel coset codes

Abstract

A new video coding paradigm (PRISM) built on distributed coding principles has recently been proposed in R. Puri and K. Ramchandran, (2002, 2003). The PRISM framework rests heavily on channel coding concepts. The basic idea is to do intra-frame coding but to approach the coding efficiency of motion-compensated inter-frame coding by using channel codes to derive a coding rate rebate by "exploiting" the correlated side-information present in the form of previous temporal frames at the decoder. The coding strategy described in R. Puri and K. Ramchandran, (2002, 2003) was limited to rate-1/2 trellis coset codes, which limit the granularity at which the correlated side-information can be exploited. In this work, we generalize the coset coding framework to multi-level codes which allow more flexible encoding rates and the capability to exploit correlations with much finer granularity. Simulation results validate the better compression performance (of the order of 1 dB for typical tests) of the multilevel framework.