Efficient HEVC Inter Prediction using SVM

In this paper, we propose a fast and efficient inter mode decision using support vector machine (SVM) for HEVC inter prediction. In the proposed algorithm, we use information such as MV variance, neighboring CU depth level and coded block flag (CBF) as features for SVM-based CU size decision; while MV variance, SKIP flag and RDO ratio for SVM-based PU mode decision. The experimental results reveal that the proposed algorithm can achieve average 29.6% reduction of computation time with 0.09% bit rate increment only, compared to the original HEVC inter prediction.