A Split-Based Algorithm Optimization for AVS3 Intra Coding

Abstract

IntraDT is an intra coding algorithm in AVS3, which is used to further divide the current coding block to obtain multiple sub blocks for prediction. The optimal prediction modes in each sub block are selected adaptively. In this paper, we propose a split-based intra coding algorithm based on the intraDT algorithm of AVS3, which makes the partition of sub blocks match the image texture more accurately. The improvement of the proposed method mainly lies in the addition of vertical and horizontal symmetric binary partition mode, and the specified transform partition modes corresponding to the two prediction partition modes, compared with intraDT. The optional prediction partition modes are extended, and the transform partition modes under the new partition modes are not necessary to be transmitted. The new partition modes can match the image texture more accurately, and find a better balance between the depth of CU partition and the redundancy of prediction block partition. The experimental results show that the proposed method is effective for the test sequences in common test condition in AVS3. The BD-rate for the components of Y, U and V achieve gains of 0.13%, 0.44% and 0.32% under All-Intra configuration, respectively, while achieving gains of 0.12%, 0.02% and 0.26% under Random-Access configuration.