Research On 3D Video Coding Method Based On Visual Perception

3D-HEVC can betterly remove the redundancy of the space-time domain and the viewpoints in the 3D video,but it can not remove the perceptual redundancy.The application of perceptual coding can reduce the coding complexity or save the coding rate while guaranteeing the subjective quality.So how to construct the visual perception model and how to apply it into 3D video coding are the hotspots of 3D video research.To solve this problem,this thesis focuses on 3D-HEVC coding standard,and studies low-complexity coding and rate-distortion optimization of 3D video from visual perception angle.According to the complexity of coding depth map,this paper applies for the Maximum Tolerated Depth Distortion(MTDD)model and proposes a fast depth map coding scheme based on virtual viewpoint synthesis for 3D video system.Firstly,it divides into four different types of depth range early decision algorithm based on the different tolerance of MTDD.Secondly,it detects the distortion of the vertical edge of the area according to the different mode selection search strategy.Lastly,it fuses these two algorithms to further reduce the complexity of the Depth video coding.Experimental results show that the proposed algorithm can reduce the encoding time by 49.45% while without obvious rendering quality degradation.Due to the huge data quantity of stereoscopic video,this paper builds a Foveated Binocular Just-Noticeable Coding Distortion(FBJNCD)model.Firstly,the effect of gradient amplitude and texture amplitude on the three-dimensional masking effect is studied by subjective experiment.At the same time,it is noticed that the Human Visual System(HVS)of visual sensitivity is not constant,the relationship between the eccentricity and visual sensitivity shows that the lager the retinal eccentricity is,the larger the pixel visual threshold is.So,we combine the perceptual characteristic of the retinal fovea of HVS and it is applied to the Multi View-High Efficiency Video Coding(MV-HEVC)test platform for asymmetric coding of stereoscopic video.Experimental results show that the proposed model can significantly save 26.04% bit rate and improve the compression efficiency of the stereoscopic video while preserving the subjective quality of the video.The traditional JND(JND)model is difficult to be applied in the stereo video.Because the JND model overestimates the visual threshold of the foreground region and underestimates the visual threshold of the background region.So this paper proposes a Stereo Just-Noticeable Distortion(SJND)model which can be applied in the stereo video.By using disparity information,the traditional JND model is divided into the foreground region and the background region,the foreground region has a smaller threshold value,and the background region has a larger threshold value.It is considering that the foreground region is more concerned with the visual center and parallax larger area.so a new significant graph is proposed based on the two rules,and different quantization parameters(QP)are given into different significant regions.The experimental results show that the proposed method can save 19.92% bit rate without compromising the quality of the stereo video.