Optimization Of Intra Prediction Algorithms In HEVC

With the development of computer processing capacities and the improvement of network bandwidth, the HD/Ultra HD video applications are being widely used. In order to compress the large-scale video more effectively, the ISO/IEC moving pictures group and the ITU-T video coding experts group together developed the Next-generation video codec standard HEVC, and completed the final draft at January2013. HEVC was designed to increase the compression performance of H.264/AVC by50%, while the final draft increased the compression rate by30%to40%.In the intra-frame prediction process of HEVC, the improvement of compression performance is brought by the35different prediction modes and the multi-scale prediction units. During the prediction process, the final prediction mode of the prediction unit is selected by rough mode decision and rate distortion optimization. The fact that all kinds of prediction units with different scales should be processed as stated above makes the intra-frame prediction part very time-consuming. Therefore, through depth research of the intra-prediction technology in HEVC, this paper designs and implements two optimized algorithms for intra-prediction according to the characteristic of high computational complexity. These two algorithms consist of a gradient-based fast prediction unit partition method and a gradient-based fast mode-decision method. Meanwhile, in the course of study, author found that the mode coding technology of the intra prediction can still be optimized. So this paper also proposes an optimized mode coding algorithm which takes the advantage of prediction mode correlation among neighboring prediction units to code the prediction mode of each unit. This algorithm can further enhance the compression ratio of the intra prediction process. In this paper, the proposed methods are tested on several standard sequences under the environment of HEVC official software HM9.0. The proposed methods are compared with the current HEVC intra-prediction algorithms. Experimental results show that all the three proposed methods are effective.