Research On Perceptual Coding Based On Spatio-temporal Influence Map

The new generation of digital video compression coding standard high efficiency video coding(HEVC)still adopts the hybrid video coding framework of predictive coding,transform coding and statistical coding.Its compression idea still stays in digital signal processing,mainly from removing time redundancy,spatial redundancy and statistical redundancy in video coding,but seldom considering redundancy in human vision.In this thesis,we extract spatial features and temporal features to obtain spatio-temporal influence map,and then apply the influence map to HEVC test model HM16.7.The main contents of this thesis include:Firstly,the perceptual features of human visual system are introduced.The spatial influence map is obtained by combining the spatial masking effect,spatial stimuli and spatial visual saliency.Global motion-compensated MV,which as temporal influence map,is obtained by subtracting the absolute motion vectors and global motion vectors.The final spatio-temporal influence map is obtained by combining the spatial influence map and the temporal influence map.According to the contents of the video sequence,this thesis validates the validity of the spatio-temporal influence map model.Secondly,the rate control algorithm in HEVC test model HM16.7 is studied and analyzed.The LCU level bit allocation algorithm is improved according to the spatio-temporal influence map model.The experimental results show that in the case of constant bit rate,the Y-PSNR of high influence region increases 0.49 dB in all intra mode,and the Y-PSNR of high influence region increases 0.21 dB in low-delay P mode.Thirdly,the rate distortion optimization algorithm in HEVC test model HM16.7 is studied and analyzed.The rate distortion optimization algorithm is improved according to the spatio-temporal influence map model.The experimental results show that in the case of constant bit rate,the Y-PSNR of high influence region increases 0.35 d B in low-delay P mode.Meanwhile,the video subjective quality is improved by about 5% with MOVIE index.