Structural based side information creation with improved matching criteria for Wyner-Ziv video coding

The Wyner-Ziv video coding (WZVC) efficiency is highly dependent on the quality of the side information (SI) created at the decoder, typically through motion compensated frame interpolation (MCFI) techniques. Since the decoder only has available some reference decoded frames, SI creation turns out to be a difficult problem in WZVC. In most the MCFI techniques available, the matching criterion used for motion estimation only takes into account the (mean of absolute) pixel differences within a block which is limiting for some content types. This paper proposes a structural based SI creation approach with improved matching criteria combining local image features, obtained from the histogram of oriented gradients (HOG), with a boundary continuity criterion and the typical pixel differences criterion. With the structural based SI creation, the motion estimation process becomes more robust, e.g. to illumination changes, thus improving the SI quality (up to 0.4 dB) and reducing the bitrate (up to 6 % in terms of the Bjontegaard metric) regarding a state-of-the-art MCFI solution.