Research On Optimization Of Hierarchical Perceptual Video Coding Algorithm

The explosive growth of video data has brought great challenges to the storage and transmission of video signals.The new generation of High Efficiency Video Coding(HEVC)standard has emerged.HEVC uses a number of advanced coding technologies to double the compression efficiency compared to the previous generation of video coding standard H.264/AVC while maintaining the same video coding quality.Signal processing-based video coding techniques are gradually approaching the upper limit of compression,and the improvement of coding performance is achieved at the cost of exponential increase of algorithm complexity.Given that the final receiver of the video signal is the human eye,scholars have taken the perceptual characteristics of the Human Visual System(HVS)to further eliminate the perceptual redundancy of the video to improve the compression efficiency.As the final receiver of the video signal,HVS dose not detect signal at the lowest visual threshold,which is called Just Noticeable Difference(JND).JND threshold can be predicted by visual perception models and can be applied to video coding to improve video perception coding efficiency.Quantization is one of the key modules in the video encoder.An appropriate Quantization Parameter(QP)can reduce the coding rate while maintaining the visual quality.HEVC uses a variance-based Adaptive Quantization(AQ)algorithm to adjust the quantization parameter offset according to the variance.The algorithm only uses the sample variance as the predictor of the visual perception model,which suffers from the problem of underestimating texture regions and overestimating distortion visibility in edge regions,so the performance of the current AQ algorithm in video encoder can be improved by deploying a better visual perception model.Meanwhile,video compression is a multi-module and multi-level collaborative coding process,including code rate control,mode selection,motion estimation,transformation,and quantization.Since video compression is a serial coding process,resulting in a rate distortion performance coupling effect for multi-module perceptual optimization,the impact of module algorithm changes on the overall coding performance does not satisfy a linear relationship.There is a lack of quantitative analysis of inter-module correlation and systematic research on multi-module joint optimization in academia,and joint optimization of multi-level perceptual coding based on inter-module correlation is still a major challenge.Aiming at the current situation of video coding research,this thesis explores the multi-module collaborative coding method based on visual perception,starting from two aspects of adaptive perceptual quantization and multi-module collaborative optimization.The research content and innovation of this thesis mainly include:1.In this thesis,JND prediction is introduced into adaptive perceptual quantization to improve the rate distortion performance of the AQ algorithm by considering the luminance adaptive masking effect and the pattern masking effect.Specifically,this thesis estimates the mean value of JND at the coding unit(CU)level through the JND threshold prediction model,maps it to the variance taking space to ensure algorithm compatibility,and derives the CU-level perceptual quantization parameters to achieve CU-level perceptual adaptive quantization.The experimental results show that the BD-BR of the adaptive perceptual quantization proposed in this thesis is reduced by 1.04% on average,and the VMAF score is increased by 0.2101 at the same code rate,which effectively improves the subjective quality of coding.2.In this thesis,a multi-level perceptual video coding optimization framework is constructed and a systematic algorithm optimization is performed on this framework.Specifically,this thesis quantitatively analyzes the correlation among the coding modules,determines the priority of each module optimization,and proposes an initial parameter selection algorithm,and simplifies the complex multi-module optimization into a continuous single-module optimization by sequentially selecting the algorithm key parameters from the discrete candidate parameter solution space.Combining the inter-module correlation and the feature parameters characterizing the image content,an online adaptive parameter model is constructed,thus realizing the multi-module cooperative perceptual coding optimization.The experimental results show that the multi-module optimization method can improve the VMAF score by 0.6936 while maintaining the same code rate.Combining the perceptual AQ optimization and multi-module optimization methods proposed in this thesis,the VMAF score is increased by 0.7609 while maintaining the same bit rate.