Effective approach to reduce complexity for HEVC intra prediction in inter frames

Recently established video compression standard HEVC has adopted quad tree based Coding Unit (CU) block partitioning structure to embrace wide variety of image texture characteristics. For spatial prediction, 35 intra prediction modes are defined in this standard to improve the coding efficiency of intra blocks. Evaluating all intra prediction modes at all depths, along with motion estimation for a block prediction, in inter frames increase the encoding complexity further. Since the possibility of intra blocks in inter frames is highly content dependent, it is computationally in-efficient to evaluate all intra modes in inter frames. To reduce the computation load of inter frames, we propose a conditional multi-stage intra prediction for inter frames. The proposed approach is a fast & pruning method with predictive nature and has two unconditional prediction stages followed by a conditional intra prediction. In this method, we categorize all intra prediction modes into few groups. Each group consists of one central angular mode with consecutive angular modes on both sides of center angle. The first stage of proposed approach evaluates all central angular modes of each group, and then the second stage evaluates two equi-distant angular modes from center angle of the winning group. Finally, the best intra cost resulted of second stage gets compared against best inter cost and then the decision will be taken to carry out further refinement of intra prediction. The performance of the proposed algorithm is verified through comparative analysis of encoding time and BD-rate.