Adaptive Wyner-Ziv Decoding Using Particle-Based Belief Propagation

In this paper we propose an adaptive Wyner-Ziv (WZ) coding scheme for non-binary correlated sources with side information at the decoder. At the encoder, a low density parity check (LDPC) code is employed to implement Slepian-Wolf (SW) coding for each bit-plane independently. The joint "source-channel" decoder, which combines SW decoding and dequantizing in a single step, preserves the symbol-domain correlation and decreases the distortion. Moreover, the incorporation of particle-based belief propagation (BP) in the decoder can extend the scheme to perform an online estimation of the correlation and enhance the BP decoding performance simultaneously. Our proposed framework is universal and can be applied to any parametric correlation model.