Side information generation using optical flow and block matching in Wyner-Ziv video coding

In this work, a new algorithm to generate high quality side information in Wyner-Ziv video coding is proposed. A block-matching algorithm is incorporated into the forward and backward optical flow and warping algorithms to find the forward and backward motion fields that are used for frame interpolation. Also, a symmetric optical flow algorithm for the purpose of frame interpolation is obtained by parameter modification in the energy functional of an optical flow algorithm. The average of the interpolated frames estimated using the forward/backward motion fields and symmetric flow is used to provide a high quality side information frame for decoding of the corresponding Wyner-Ziv frame in the Wyner-Ziv video coding problem. The proposed algorithm significantly improves the quality of the side information frames compared with those provided by the advanced block matching frame interpolation in the typical Wyner-Ziv video codecs. Simulation results showing significant improvements in side information quality and rate-distortion performance in Wyner-Ziv video coding are provided.