Advanced Correlation Noise Model for Pixel - Domain Wyner - ziv Video Coding ⋆

The capability of the correlation noise model (CNM) to describe the noise distribution between the original frame and its side information (SI) plays a notable role in the coding efficiency of Wyner-Ziv video coding (WZVC) paradigm. One of the problems of the most available Laplacian-based CNM is that it may fail for high motion sequences or long group of picture (GOP) sizes. To solve this problem, this paper proposed a novel CNM for pixel-domain WZVC which able to adapt to video content and coding parameters by modeling the case of high SI quality as Laplacian distribution while modeling the case of low SI quality as Cauchy distribution. In addition, the parameter estimation for Cauchy distribution is based on quantile estimators. Experimental results show that significant rate distortion (RD) improvements of the proposed novel CNM are achieved especially for high motion video sequences and large GOP sizes.