Side Information Generation for Distributed Video Coding

Side information (SI) generation is one of the key components of a Wyner-Ziv coder. In this paper we present a novel multi-frame SI generation approach which uses adaptive temporal filtering to estimate the pixel values for SI and motion vector filtering for refinement. For temporal filtering, we derive the optimal mean squared error temporal filter when the noise can be evaluated, and propose a similarity weighted temporal filter when the knowledge of the noise is not available. The temporal filter adapts on the quality of the motion estimation. The quality of SI generation is further improved by using motion vector filtering to reduce the noise effect from motion estimation. Experimental results indicate that the proposed SI generation approach yields good performance in terms of SI quality and conditional entropy.