Research On HEVC Intra - Frame Fast Mode Decision And Depth Judgment

With the increasing requirements for video quality, HD video and ultra HD video is getting more and more popular. Therefore, the most widely used video coding standards represented by the H.264/AVC are becoming increasingly difficult to meet people’s needs. So the JCT-VC (Joint Collaborative Team on Video Coding) organization founded by VCEG and MPEG developed the latest video coding standard:HEVC (High Efficiency Video Coding).HEVC continues to use the H.264/AVC video coding framework which includes prediction coding, entropy coding and block partition. In intra coding, there are 35 intra prediction modes in HEVC, which are more than the 9 in H.264/AVC. Compared with the 16 X16 size macroblock in H.264, HEVC uses the quadtree partition algorithm and the biggest block size is 64 X 64. But the attendant is the increasing complexity of encoding.In order to compress video, intra prediction can remove the spatial redundancy by exploiting the strong correlation between adjacent pixels of the image. This paper further analyzes spatial characteristics of coding units to improve the prediction efficiency in two aspects. On the one hand, the directional characteristics of the local-edge of the image can be used to reduce the number of intra prediction modes. On the other hand, if it needs smaller coding units can be judged through the analysis of the coding unit. So it can reduce the number of coding units to be predicted to achieve the purpose of improving prediction efficiency. In the aspect of mode selection, this paper use Sobel operators to detect the texture of prediction units, and establishes the weighted histogram of each mode, proposes a fast intra prediction algorithm based on local-edge. Compared with HM11.0, the experiments show that this algorithm can save about 29% encoding time on average while keeping the quality of video. In the aspect of coding depth decision, this paper stops some unnecessary sub block partition by analyzing the coding unit is flat or has a direct local-edge. Experiments show that this algorithm can save about half of the intra coding time and keep the quality of the video basically unchanged.