A generalized correlation noise model for pixel domain Wyner-Ziv video coding

Wyner-Ziv video coding (WZVC) is a video compression paradigm based on distributed source coding principles. WZVC allows reducing the complexity of video processing at the encoder thus providing lower energy consumption. One of the key operations of WZVC is correlation noise modeling which is responsible for estimation of statistical characteristics of errors arising in virtual dependency channel. In this paper a new generalized correlation noise model based on Hidden Markov Random Fields (HMRF) is proposed. The model considers spatial characteristics of errors and uncertainty of noise estimation in an explicit form via probabilistic description of noise using Markov Random Fields. The approach introduced in this paper allows considering the problem of correlation noise modeling in the context of maximum á posteriori (MAP) inference in HMRF framework.