| Although 3D video can provide immersive visual experience,but also bring enormous pressure of video storage and transmission because of its big data.Therefore,under the premise of ensuring the subjective quality of video,as much as possible to occupy a smaller transmission bandwidth is major challenges be solved in 3D video coding field.Previous coding algorithms have focused on removing spatial and temporal redundancy,but we have turned our attention to removing visual redundancy in the video.Thus,this paper proposes to improve the compression efficiency by combining the human visual characteristics with the existing H.264 coding framework,and aiming at ensuring the subjective quality and reducing the code rate as much as possible.Region of Interest(ROI)coding is to improve the compression efficiency while ensuring the subjective quality of the video by controlling the allocation of the macroblock quantization parameter(QP)of the video background region and the ROI region.For 2D video,detection performance of ROI is unstable,which limits the promotion and use of ROI coding.The depth information contained in the 3D video is closely related to the degree of interest in the human visual system(HVS),which provides favorable conditions for ROI detection of 3D video.Therefore,this paper presents two algorithms for 3D saliency detection based on the depth information.Just Noticeable Distortion(JND)refers to the phenomenon that different regions of the image have different distortion sensitivity caused by the physiological characteristics and psychological characteristics of the human visual system.When the degree of distortion of the specific region is below the JND threshold,the human eye cannot perceive its presence.JND video coding technology is mainly aimed at removing the visual redundancy of video,with the help of a reasonable allocation of coding resources.In this paper,ROI and JND coding compatible with the H.264 standard are studied in 3D video(single-view video and depth map format).This paper first studies the human eye stereo vision system model,and analyzes the relationship between the depth of the scene and the degree of interest in the HVS.In this paper,a depth-based stereo projection saliency detection algorithm is proposed,and the relationship between the depth and the background area is further explored.A 3D saliency detection algorithm based on background detection is proposed.Aiming at the different attention degree of the human eye to different depths and the mutual occlusion of objects between different viewpoints,a model is proposed to calculate the JND threshold using the depth map.After that,the paper discusses the composing and adjusting method of H.264 compression standard and compressed bit rate.According to this,before the video is compressed by H.264,the region of the video frame is divided by ROI and JND,and the quantization model of human visual system is constructed to guide the selection of quantization parameters of ROI.At last,the influence of the quantization strategy on the bit rate after video compression is analyzed both theoretically and experimentally.The experimental results show that the proposed scheme can preserve the human visual sensory region with low distortion at the same bit rate,and provide a better visual experience for the user. |
