Research And Optimization Of H.266/VVC Intra Prediction Algorithm

The Joint Video Exploration Team released the newest Versatile Video Coding(H.266/VVC)standard.Compared with the previous High Efficiency Video Coding standard,VVC introduces a novel quadtree with a nested multi-type tree partition structure.The more flexible coding unit(CU)size adapts to the video better and improve the compressing performance.However,it causes tremendous computational complexity to VVC at the same time.It is necessary to optimize the coding process to promote the coding efficiency of VVC standard.In this thesis,two optimization algorithms are proposed for CU partition of VVC intra prediction:(1)Considering that the final partition structure of VVC intra coding is highly correlated with CU texture,a low-complexity CU partition scheme based on the texture characteristic is designed.Specifically,the texture complexities of vertical and horizontal directions are quantitatively measured in terms of the sum of the mean absolute deviation(SMAD)of the subblocks.And then,the vertical and horizontal texture complexities are compared to eliminate some unlikely multi-type tree(MT)partition modes and reduce the encoding complexity.In addition,the threshold of directional SMAD ratios for choosing the texture direction is adjusted according to the quantization parameter to improve the prediction accuracy.(2)A fast CU partition algorithm based on spatial correlation is proposed.This algorithm is based on the similarity between the current blocks and neighboring blocks within the same frame.The best intra prediction modes of the left and above CU are referenced to measure their texture characteristics.The partition of CUs whose neighboring blocks are flat content is terminated early.Then,according to the texture direction of neighboring blocks,the unlikely MT partition mode of current CU is excluded to achieve fast CU partitioning.Simulation experiments are carried out for the above two optimized intra prediction algorithms on various resolution sequences respectively.Experimental results demonstrate that the first proposed method reduces 30.33%coding time on average than the original VVC test model at negligible cost of 0.57%Bj?ntegaard Delta Bit Rate(BDBR)increase and 0.03dB Bj?ntegaard Delta Peak Signal-to-Noise Rate(BDPSNR)decrease.The second method achieves 17.83%time saving at negligible cost of 0.61%BDBR increase and 0.03 PSNR decrease.It can be seen that the algorithms proposed in this thesis can alleviate the computational load of intra prediction effectively with negligible coding loss.