Modeling non-stationary correlation noise statistics for Wyner-Ziv video coding

Distributed video coding is a new video coding paradigm, with the ability to shift coding complexity from encoder to decoder. Turbo code based pixel-domain Wyner-Ziv codec is one of the most practical lossy DVC schemes. The performance of Wyner-Ziv codec significantly depends on the modeling of correlation noise statistics between the side information and the original Wyner-Ziv frames. It was observed that the correlation noise statistics differs temporally and spatially. In this paper, a novel algorithm is proposed to model the non-stationary correlation noise statistics. The experiments show that the proposed algorithm has led to significant improvement for Wyner-Ziv coding performance.