Research On Perceptual Video Compression Based On Intra Secondary Prediction

With the emergence of various emerging video applications and ultra-high resolution video,the amount of video data in the world is growing rapidly.A large amount of video data brings challenges to the process of video processing,storage and transmission.Video coding technology is one of the key solutions to the above challenges.For a long time,the international standards organization has been committed to developing more efficient video coding standards.Prediction,transformation,quantization,entropy coding and other technologies are used in video coding to improve its compression efficiency.Efficient video coding technology makes the space occupied by video greatly reduced,and then can be stored and transmitted conveniently,which is the key to ensure the quality of service and user experience of video applications.Therefore,how to improve the compression efficiency of video coding is a very popular and significant research topic.The mainstream video coding frameworks usually use differential prediction coding to obtain the prediction residual,and transform,quantize and entropy encode the residual data to achieve the purpose of video compression.Prediction coding is divided into intra prediction and inter prediction,which are used to eliminate spatial and temporal redundancy in video respectively.Transform and quantization can reduce the high frequency signal which is not sensitive to human eyes in the video,while entropy coding is used to eliminate the statistical correlation between the syntax elements in the coding process.In addition,with the research of human visual characteristics,the perceptual video compression technology for subjective visual quality has been widely used.Perceptual video compression uses the human visual model to calculate the perceptual redundancy in the video,and improves the compression efficiency by removing these redundant information.In this paper,we will study the perceptual video compression based on the intra secondary prediction algorithm to further improve the compression efficiency.The reference pixels of traditional intra prediction are located in the left and above part of the coding block,and the right and bottom part of the coding block is far away from the reference pixels,which means that their spatial correlation is poor,so the prediction accuracy of these pixels is low,and the final prediction residual is large.In order to solve this problem,this paper proposes an intra secondary prediction algorithm.Based on the traditional intra prediction,the reconstruction pixels of the right and bottom boundary of the coding block are used as new reference pixels,and the left and above reference pixels of the coding block are used to predict the coding block again,which improves the prediction accuracy and reduces the prediction residual.On the basis of intra secondary prediction,the residual can be further reduced.Traditional coding technologies usually use the objective difference between the image before and after coding as the standard of coding quality evaluation.But the final receiver of video is human eyes,so the result of objective coding is difficult to meet the subjective feeling of human eyes.In order to solve this problem,we will use the luminance adaptive effect and contrast masking effect of human eyes to calculate the just noticeable distortion in the image,and design a new residual filter to filter the perceptual redundancy in the residual.At the same time,considering the influence of quantization distortion in the filtering process,this paper proposes a distortion estimation model suitable for the new generation of video coding standard VVC(versatile video coding),and calculates the quantization distortion in the coding process to control the precision of residual filtering and ensure the subjective quality of the coding image.Through the perceptual residual filtering algorithm based on the intra secondary prediction in this paper,the residual data in the coding process is greatly reduced,which saves the bits needed for the coding residual,ultimately reduces the coding rate and improves the compression efficiency of video coding.Experimental data shows that,compared with the new generation of video coding standard VVC,the algorithm of intra secondary prediction and perceptual residual filtering can save 1.18% and 5.49% of the coding rate on average,and the algorithm of merging the two can save 6.5% of the coding rate on average,and obtain the similar subjective coding image quality.