Fast mode decision algorithm for AVS3 intra prediction

Abstract

AVS3 is the latest generation of video coding standard developed independently by China. It obtains 50% bitrate savings under the same visual quality as compared with the previous generation coding standard AVS2, largely due to availability of numerous new coding tools. However, these newly introduced coding tools increase the computational complexity significantly and thus severely hinder the real applications of AVS3. In light of the above, this paper presents an innovative fast intra mode decision method that exploits texture information. In particular, for each Prediction Unit (PU), the Sobel operator is employed to obtain the gradient information, which helps to produce candidate intra prediction modes. In addition, to trade off coding efficiency against computational complexity, the PU content flatness is exploited to determine the number of candidate modes. The proposed method is implemented on HPM-4.0.1, and the experimental results demonstrate that it obtains 35.2% encoding time saving with moderate compression performance loss.