Research Of Fast Algorithm Of Intra Prediction For HEVC

With the development of multimedia technology and the expansion of network bandwidth, people’s request for video quality is higher and higher. Then the high definition videos whose resolutions are more than 2K and 4K emerged. H.264/AVC, which is the most common video coding standard, can’t provide satisfactory compressibility for these videos with high resolution because of the small size of its macro-block. To solve this problem, H.265/HEVC (High Efficiency Video Coding), which is the newest video coding standard, was published by JCT-VC (Joint Collaborative Team on Video Coding). The aim of HEVC (High Efficiency Video Coding), which is the newest video coding standard, is to sharply raised the efficiency of video coding. With the same quality of image, the compression ratio of HEVC will be twice as that of H.264/AVC High Profile.In order to achieve this goal, many parts of the encoding process is improved. They include coding tree unit, motion prediction, entropy coding and loop-filter etc. What is about intra prediction is that the number of luma prediction modes is increased from 8 to 35 and the max size of coding unit and transform unit is expanded from 8x8 to 64×64 and 32x32. Because of prediction modes and more depths, the precision of prediction in HEVC is improved. But also because of it, the complexity of calculation of HEVC is raised.This paper introduces coding unit and the process of intra encoding in detail. Following, the influence from partitioning depths of CU to the rate-distortion performance and calculation of encoder is tested. Then an overview of the fast intra encoding algorithms which have been proposed by others is given.The algorithm proposed in this paper decreases the calculation of intra prediction for HEVC by reducing the amount of modes and depths which should be traversed. In one depth, this paper reduces the amount of prediction modes from 35 to about 21 by interlaced extracting modes for RMD (Rough Modes Decision); also based on the theory of spacial similarity, reduces the amount of modes to about 3 by statistics of pixel gradient and comparing modes of adjacent PUs for RDO (Rate-Distortion Optimization). Besides, when completing the prediction of a PU in which depth isn’t 0, the richness of texture information of the current CU can be calculated by comparing the total rate-distortion cost of PUs in the same depth and the threshold gained from calculating the SATD (Sum of Absolute Transformed Difference) of the sub-PUs of the PU in previous depth. Based on it, the depth traversing of the current CU can be early terminated. The test results shows that, if using this combined algorithm, the encoding time of the encoder can be saved, and the increasement of bitrate and the decreasement of reconstructed images are all slight.