SKIP mode applicable motion vector prediction for high-performance video coding

Abstract

In hybrid video coding, motion vector prediction (MVP) plays an important role for initial motion estimation search points, SKIP mode motion vector (MV) and reducing amounts of bits to transmit the motion vector information. More accurate prediction for the motion vectors found in motion estimation process produces fewer amounts of bit-rates for transmission or storage. Therefore, research for accurate MVP is needed to enhance coding efficiency. The recently standardized H.264|MPEG-4 Part 10 Advanced Video Coding (AVC) uses as a motion vector predictor the median of three motion vectors in the spatial neighbor blocks of the current block to be inter-predictive coded. Although this median predictor is well suited at global motion, it exhibits poor prediction performance when motion is complex. In thesis, we design a new MV predictor set starting with all combinations of the horizontal and vertical components of the neighboring MVs for motion vector prediction of skip mode. The most eight probable combinations of the horizontal and vertical components of the neighboring MVs are defined as our new motion vector predictor. This leads to enhance the chance of skip mode selection so that coding efficiency is improved. The experimental results show that the proposed method reduces 6% of bitrates in BDBR compared with the media predictor of AVC.