Maximum likelihood estimation for the generation of side information in distributed video coding

The unique requirements imposed by contemporary video communication applications promote the rise of the distributed video coding paradigm, which offers flexible complexity allocation and error resilience combined with competitive compression performance. This work proposes a novel Wyner-Ziv video coding scheme, performing hash-based motion estimation at the decoder. Subsequently, maximum likelihood estimation generates side information from a collection of candidate predictor values. Moreover, a supplemental Wyner-Ziv enhancement layer is coded in the transform domain. The presented Wyner-Ziv coding scheme delivers state-of-the-art compression performance, achieving average Bj-ntegaard rate savings of up to 9.31% compared to a state-of-the-art reference codec.