Dissertation > Industrial Technology > Radio electronics, telecommunications technology > Communicate > Image communication, multimedia communication > Image coding

Research on Optimization and Coding Technology of Depth Map in3D Video

Author ZhangJunTao
Tutor LiBo; YuLi
School Huazhong University of Science and Technology
Course Communication and Information System
Keywords 3D video Kinect Depth map optimization Depth coding Depth mapup-sampling View rendering
CLC TN919.81
Type Master's thesis
Year 2013
Downloads 13
Quotes 0
Download Dissertation

Compared with texture image, depth map can be saved coding rate and easilyrendered virtual viewpoint by Depth Image Based Rendering, texture plus depth ofmulti-view video is considered to be the most effective3D video representation, depthinformation is indispensable important information for3D video data expression.Compared with general texture image, the depth map is object interior smoothing and edgeoutstanding characteristic, and isn’t used to the output display and looked, only to drawnew viewpoint. Due to the differences between the general2D video and depth video, it isnot rational to directly adopt the traditional video way for depth processing, so wepropose new optimization model of depth map and efficient coding method making fulluse of characteristics of depth map.First of all, we propose an effective optimization model of depth map, which notonly removes noise but also maintain clear object boundaries, and at the same time,considering the four measures against their neighbor pixels in color and depth image, thevalues of pixel position hole regions can be recovered.Secondly, we propose Edge-directed interpolation for down/up-based depth map.Preservation the edges in depth map coding is important for improving synthesized viewquality, this paper presents a novel edge-preserving depth up-sampling method fordown/up sampling-based depth coding using both the texture and depth information. Thedepth up-sampling algorithm took into account the edge similarity between depth map andcorresponding texture image, and structural similarity between low resolution depth mapand high resolution depth map. Based on these characters, the optimal minimum meansquared error (MMSE) interpolation coefficients were estimated from the local covariancecoefficients of down-sampled depth map and corresponding texture image.And lastly, we propose depth map coding based on visual perception. As theperformance of depth map coding mainly depends on the subjective and objective qualityof rendering view, in the depth map coding, not only should we consider reducingrendering distortion of the virtual view, but we should keep consistence between distortionmeasure and subjective opinions. We build an estimation model of rendering distortion ofthe virtual view which is caused by depth value change that is produced by depth coding.The measure of distortion is based on image quality evaluation method SSIM that is basedon human visual system model. Besides, we put the estimation model in the RDO which isbased on visual perception SSIM. When coding depth map, what we built-RDO model can improve objective (SSIM measure) and subjective quality of the rendering virtualviewpoint quality, by means of choosing optimal macro block coding mode (depth mapmacro block coding mode when virtual view distortion is minimum).In a word, by making full use of characteristics of depth map, we propose newoptimization model of depth map and efficient coding method. Complete accurate depthcan be obtain by the depth map optimization method, and then the proposed codingmethod for depth map compression coding can get higher coding efficiency

Related Dissertations
More Dissertations