New merge mode decision in High Efficiency Video Coding (HEVC)

摘要

The advanced motion vector prediction (AMVP) used in the newly developed High Efficiency Video Coding (HEVC) employs five MV candidates from both spatial and temporal domains for merge mode decision. The merge index is then entropy coded using a simple prefix code. To reduce merge index overhead, in this paper we propose a new merge mode decision algorithm which suggests using decoder derived MV candidates for merge mode decision, based upon both sum of absolute bidirectional prediction differences (SABPD) and template matching (TM) criteria. As a result, only a merge index flag is required for transmission instead of prefix coded index. The experimental results reveal that the proposed algorithm achieves up to 1.46% of bit rate reduction (with average 0.73% reduction), compared to the method used in the HEVC.