Research On Video Coding Based On Visual Perception

The development of modern communication technology makes remote education and virtual video conferencing,wireless video communication technology and digital video surveillance service enter into the people's field of vision.These high-end multimedia services have also created a huge video capacity while presenting high quality visual experience,storing and transmitting this high capacity video information,and put forward higher requirements for video compression efficiency.The efficient video coding HEVC(High Efficiency Video Coding)usually uses objective distortion measurements to optimize the algorithm,but these methods do not take full account of the visual perception characteristics.Therefore,the combination of human visual perception and video coding can greatly improve the efficiency of video coding,improve the rate distortion performance of video coding and optimize the subjective quality of video.In view of this,this studies the key technology of HEVC video coding based on human visual perception,proposes a rate distortion optimization method based on structural similarity,and establishes a coding model based on visual perception characteristicsand a visual perception coding optimization algorithm based on visual significant graph.First,the distortion metric similarity SSIM(structural similarity index measurement)which reflects the characteristics of video structure is studied A rate distortion optimization method based on structural similarity is proposed.This method uses 1-SSM as an algorithm for distortion measurement.It uses the relationship between the mean square error and the structure similarity to improve the structure similarity,and uses the improved structure similarity to modify the rate distortion function and adjust the rate distortion function.The Lagrange factor is added to achieve rate distortion optimization.The experimental results show that,compared with the proposed algorithm and HM algorithm,the average bit rate is reduced by 2.9% for the same structure similarity,and the average rate of rate distortion is only 0.8% for the same PSNR.The result shows that the proposed structure similarity algorithm can improve the rate loss ability of video coding.Secondly,in view of the shortcomings that the structure similarity can only reflect the video structure information,proposes a HEVC coding model based on visual perception,which combines the subjective visual characteristics of the human eye.The model uses the GBVS(Graph-Based Visual Saliency)algorithm to establish the visual saliency model,and through the algorithm,the image feature information is extracted and the video significant map is obtained,and the video image is divided into significant region.Domain and background area,and extract significant weight of video.Finally,according to the proposed visual salience model and the problem of unreasonable rate allocation in the HM model,proposes aconspicuous coding tree unit rate allocation method,which adds significant weight to the complexity weight.When the target bit rate is large,both of them guide the code rate allocation at the same time.When the target bit rate is small,it only uses significant weights to guide the bit rate.The allocation is aimed at keeping the target code rate unchanged.A significant weight is added to guide the allocated Lagrange factor of the rate adjusted coding tree unit,and the Lagrange factor and the quantization parameter are corrected.At the same time,the adjusted Lagrange factor is used to modify the rate distortion function of the coding unit to make the coding unit of the video significant region select the optimal encoding mode,thus improving the coding effect of the significant region coding unit module.The experimental results show that,compared with the official HM16.8 algorithm,the Y-PSNR in the significant region rises by0.47 d B;RA mode,the Y-PSNR in the significant area increased by0.51 d B,the rate distortion performance of the significant area of video coding has been significantly improved.