Fast H.266 Intra Prediction Algorithm Based on Structure Tensor

The new generation of video coding standard H.266 expands the intra prediction modes from 35 to 67, though improves the prediction accuracy, the coding computational complexity increases a lot. According to the high correlation between the direction of video texture and the prediction mode decision, a fast H.266 intra prediction mode selection algorithm based on structure tensor is proposesd in this paper. For each coding unit CU, a structure tensor matrix is calculated, and then, eigenvectors obtained by singular value decomposition are utilized to predicte the direction of the texture and narrow down the range of candidate modes, and reduce the number of modes for rate distortion cost calculation. Experimental results show that under all-intra (AI) configuration, the proposed algorithm achieves 17.28% coding time saving with only 1.12% bitrate increasement compared with H.266 VTM-7.3 standard platform, which reduces the coding complexity effectively.