Research And Implementation Of Intra Coding Optimization For AVS3

Video coding is a technique that compresses the original video data into bit stream.It can reduces video data with the requirements of certain video quality.With the development of ultra high definition(UHD)and virtual reality,the amount of video data is increasing rapidly,which brings pressure to the transmission and storage.In order to adapt to the growing amount of video data and the diversified video content,the new generation of video coding standard AVS3 is already in full swing.AVS3 incorporates lots of new tools on the basis of AVS2 standard,and achieves higher compression ratio while keeping the video quality consistent with AVS2.However,it also leads to higher coding complexity.In order to reduce the coding complexity and speed up the coding time,the optimization of intra coding in AVS3 is studied in this thesis.The fast algorithm based on intra prediction filtering and block partition are proposed.The specific research of this paper is as follows:1.In order to accelerate the time of intra coding,the complexity of AVS3 intra coding is analyzed,and a fast intra coding algorithm based on intra prediction filter(IPF)is proposed in this thesis.In the process of intra prediction,there is a correlation between results of two rough mode decision(RMD)respectively.Based on this,the algorithm utilizes the results of first RMD process(IPF off)to reduce the candidates in second RMD process(IPF on).In the RMD process,we record a subset modes from the entire intra mode candidates with least Hadamard costs.In the second RMD process,we only check these selected modes.Experimental results show that the proposed method can achieve 16%and 10%of time saving for all intra configurations in two typical AVS3 software,HPM4.0 and SVT-AVS3.The BD-rate Y increasing is negligible with only 0.06%and 0.03%respectively.2.To reduce the number of block partition attempts,this thesis proposed two fast algorithms.The first algorithm is based on decision tree and the second algorithm is based on the best intra mode in prediction of current coding unit.For the first algorithm,the texture of current block and sub block and the results of intra prediction in current block are used as features to train two kinds of offline models.One is used to judge whether or not to try BT(binary tree)partition,and the other is used to judge whether BT is divided horizontally or vertically.each size of the block is trained separately,and only the block sizes with good training performance is used in this algorithm.Experimental results show that the proposed method can reduce the coding time by 17%,while the BD-rate Y increases by 0.81%.Based on the statistical results,mode 12 and 24 are optimized in the second algorithm.When the best intra mode is 12,the horizontal BT partition will not be allowed in current block.When the best intra mode is 24,the vertical BT partition will not be allowed.Experimental results show that the proposed method can reduce the coding time by 14%,while the BD-rate Y is increased by 0.41%.The algorithm can also achieve 12%coding time reduction with 0.21%BD-rate Y when only mode 24 is optimized.