The optimization of side information generation algorithm for Wyner-Ziv video coding

Abstract

In order to improve the quality of the reconstructed Wyner-Ziv frame, based on the introduction of a basic way to generate the side information, two strategies was proposed to optimize the side information. Firstly, during the unidirectional motion estimation process, we made full use of intra-correlation and inter-frame correlation to adaptively obtain the motion search starting coordinates of the current macro-block and the size of the search window. After the bidirectional motion estimation, a motion vector smoothing method was introduced to optimize the reconstruction of the motion vector field, and to further enhance the authenticity of the motion vectors. Compared with the existing conventional interpolation algorithm, the experimental results show that the average PSNR of side information generated by the proposed method can be improved by 1.33~1.66dB, and the subjective quality of the side information is significantly improved.