Geometric transforms and reference picture list optimization for efficient disparity compensation

This paper presents a method to increase the efficiency of the disparity compensation process performed by state-of-the-art video coding extensions for 3D and multiview. Disparity compensation relies traditionally on a block matching algorithm limited to two degrees of freedom, that in many cases does not adequately address the disparity created by the camera arrays. The proposed method is combined with MV-HEVC in order to improve the efficiency of inter-frame prediction, by generating additional reference pictures, using 3D scene information. The new geometrically compensated pictures are chosen and reordered based on their usefulness for prediction and associated cost in the encoding process. When compared to MV-HEVC, the proposed scheme is able to achieve bitrate savings of up to 3.19% for convergent camera arrangements and 2.41% for parallel camera arrangements.