The Research On Fast Intra-prediction Algorithms For AVS3

AVS(audio video coding standard)is a series of video and audio coding standards with independent intellectual property rights.In order to meet the needs of the development of the video and audio industry,the AVS working group has developed a new generation of video and audio coding standards,referred to as AVS3,since 2018.At present,the video part of AVS3 has completed the first phase of task,"information technology intelligent media coding part 2: video" was first released as a group standard on May 13,2020,with the standard number of T / AI one hundred and nine point two-2020.AVS3 adopts many new technologies,which greatly improves the coding efficiency and increases the coding complexity.Therefore,the research of AVS3 fast coding technology is very important.In this paper,the fast intra prediction coding technology of AVS3 is studied as follows:Aiming at the problem that new block division structure of AVS3 makes the coding more complex,this paper takes the current process of coding unit(CU)division as the classification problem modeling,introduces the decision tree for fast CU division,extracts the eigenvalue information according to the image texture features and spatial neighborhood correlation,uses the Gini coefficient as the tree node judgment basis to train the decision tree.The accuracy of the model is 80.19%.Then the decision tree is integrated into the official test software to replace the high complexity calculation of the original deep recursive call of CU.Finally,the performance of the algorithm is tested,and the effectiveness of the algorithm is verified from the aspects of objective quality and subjective quality.The experimental results show that the proposed algorithm reduces the coding time by 81.97% on average,and the average loss of BDPSNR(Y)is 0.34 d B.In the development of real-time encoder,speed is the primary consideration,so this algorithm has a good reference value.In addition,the directional intra prediction mode of AVS is studied deeply,and the calculation methods of 30 directional intra prediction modes are analyzed.It is found that in the process of interpolating filter to predict pixel value,because the filter coefficient is periodic,many pixel points use the same reference pixel and filter coefficient,so the calculated prediction value is the same.Based on this finding,in order to reduce the computation time of coding,this paper proposes to directly copy the calculated pixel values,and the conditions for each mode to copy pixels are different.A large number of experimental results show that the optimized coding time can be reduced by about 4 times on average.And this method will not bring any loss of compression efficiency.