Research On AVS3 Video Coding Algorithm Based On Video Features

With its intuitive viewing experience and rich content,video has gradually become the main channel and way for people to obtain information and entertainment.At the same time,the development of video is gradually moving towards HD and UHD.The surge of video data volume and the improvement of video quality bring great opportunities and challenges to video transmission and storage.Relevant institutions at home and abroad have been scrambling to carry out research on the new generation of video compression coding standards one after another.The third generation standard AVS3 formulated by the domestic AVS working group has entered the final stage.A large number of new coding technologies have been introduced into the AVS3 standard,which greatly improves the coding performance but also increases the coding complexity sharply,which is not conducive to the popularization and application of the standard.In order to reduce the coding complexity,this paper deeply studies the characteristics of DCT coefficients of the video image from the perspective of the frequency domain of the video image,and designs the relevant optimization algorithm for the AVS3 intra-frame prediction coding module.The specific work is as follows:1.In the selection process of the intra-angle prediction mode,using the image texture characteristics reflected by the low-frequency coefficients of the frequency-domain DCT,this paper proposes an algorithm to reduce the number of traversal modes in the RMD process of the intra-angle prediction mode.The algorithm firstly divides the 18 prediction modes in the first step in the RMD process into 6 groups according to the angle.Then,the approximate texture angle of the block to be predicted is calculated by using the DCT transform low-frequency coefficients.Then,the angle mode of the RMD process is initialized according to the interval in which the calculated angle falls,so as to reduce the angle mode to be traversed and reduce the coding complexity.Among the three designed exploration schemes,the optimal scheme brings an average coding time saving of 42%in the rare case that the Y component BD-rate loses 0.66%on average.2.For the selection of intra derived tree,this paper proposes an optimization algorithm that skips unnecessary intra derived modes in advance.The algorithm calculates the texture angle of the current block by using the method of calculating the texture angle of the image block in the intra angle prediction mode selection optimization algorithm mentioned above,and then skips the three intra derived modes that are inconsistent with the texture angle direction of the current image block.Then,the current block is divided in half,and the DC coefficient variances of the sub-blocks in the two blocks are calculated respectively to determine the similarity of pixel features in the block,and then one of the two asymmetric divisions is skipped.The last remaining modes are determined by the original code algorithm to determine the best intra derived mode.The skip ahead in this algorithm is more explicit,and the DCT transform coefficient is used to make the decision instead of the process of RD cost in the original algorithm,which saves the coding time.Experimental results show that the algorithm saves 28%of encoding time with an average performance loss of only 0.16%.