Research On CTU Layer Rate Control Algorithms Of HEVC Based On Visual Perception Characteristics

High Efficiency Video Coding(HEVC)is the inheritance and development of H.264/AVC,which further improves video coding efficiency while maintaining good video quality.HEVC adds or improves new coding techniques in almost every module.The efficiency of these modules with new coding techniques makes HEVC coding performance greatly improved.In the case of similar compression quality,HEVC can save about 50% of the code rate.However,the HEVC standard compresses video mainly when it considers time redundancy and spatial redundancy while ignoring a large amount of visual redundancy.This thesis obtains the final saliency map by combining temporal and spatial saliency information and applies it to vision-based perceptual coding.The work of this thesis includes the following aspects:(1)A scheme combining the spatial saliency feature and temporal saliency feature to the final saliency map is proposed.The saliency weight of each Coding Tree Unit(CTU)is obtained according to the saliency map,and then adjusting the quantization parameter of the CTU layer according to the significance weight value.The quantization parameter(QP)of the significant region is reduced,the distortion is smaller,and QP of the non-significant region CTU such as the background is increased,making the subjective quality of the entire frame increase with limited bits.In this thesis,the significant weighted peak signal-to-noise ratio(Saliency Weight Peak Signal Noise Ratio,SW-PSNR)is used to evaluate the subjective quality of video coding.From the experimental results,it can be concluded that compared with the existing algorithm of HM16.7,the SW-PSNR is improved by about 0.27 dB in the all Intra mode.The Y-PSNR of the significant region is increased by about 0.28 dB on average,and the SW-PSNR is improved by about 0.24 dB on average in the coding mode of the lowdelay IPPP.The Y-PSNR of the significant region is increased by about 0.28 dB on average.(2)A distortion metric based on visual saliency is proposed.The visual characteristics of human eye are taken into account in the relationship model between Lagrangian factor and distortion,and a relation model between Lagrangian factor and perceptual distortion is obtained,so that the objective distortion metric is consistent with the subjective perception of the viewer.In addition,we comprehensively consider the bit allocation and actual consumption information of the CTU of the same frame in the previous frame and the previous frame to adjust the Lagrangian factor of the CTU of the current coded frame,and finally improve the subjective quality of video coding with limited bit resources.The experimental results show that compared with the existing algorithm of HM16.7,the Y-PSNR of the salient region in the video is improved by about 0.16 dB in the low-delay IPPP coding mode,which is higher than the Wang's algorithm about 0.10 dB.It can be concluded from the experimental results that the proposed algorithm based on visual saliency adjustment of CTU layer QP and the sensibility-based perceptual distortion algorithm greatly improve the subjective quality of video coding under the condition of ensuring similar code rates.