A probabilistic predictor for side information generation in distributed video coding
 
A probabilistic predictor for side information generation in distributed video coding 
 
Frederik Verbist, Nikos Deligiannis, Marc Jacobs, Joeri Barbarien, Peter Schelkens, Adrian Munteanu, Jan Cornelis
 
Abstract 

In distributed video coding, the quality of the side information is of critical importance for the compression performance of the entire system. This paper proposes a novel transform domain Wyner-Ziv video coding scheme performing hash-based overlapped block motion estimation at the decoder. In this framework, a novel method to optimally generate side information from a collection of side information candidates is presented. By adopting the side information dependent correlation noise paradigm and exploiting the information contained in the hash, average Bj{\o}ntegaard rate savings of up to 22.26% compared to previous techniques are reported. Moreover, the presented Wyner-Ziv coding scheme, incorporating the proposed side-information generation technique, achieves average Bj{\o}ntegaard rate savings of up to 11.18% compared to the state-of-the-art DISCOVER codec.