Research On Key Technologies For Three-Dimensional Video Coding

The3D video (3DV) can provide the audience a more vivid word and realperception, thus it attracts widely concerns and researches. The raw data of3DV istremendous (the main data format is multi-view plus depth), which can hardly betransmitted by the current network. Thus, the compression of the original data, whichcan simultaneously adapts the load of the current network, and guarantees the visualperception of the3D video, is a quite challenging task. Currently, a special team on3DV coding was established by the Moving Picture Experts Group (MPEG) to study on3DV coding. Meanwhile, a lot of references also focus on this research area.In this context, this paper mainly concerns about the coding tenures of3D video,and contributes the following achievement:1．An efficient depth image intra-frame coding method based on the dependency ofthe video and the depth image is proposed to reduce the complexity of the depth imageintra-frame coding. In the proposed method, the optimum intra-frame coding mode ofvideo is used to predict the optimum intra-frame mode of depth image coding in orderto simplify the process of Rate-Distortion Optimization (RDO) and reduce thecomputation of Lagrangian cost function in intra-frame prediction in depth imagecoding. Simulation results show that the proposed method can reduce the coding timeby37.16%in average compared with the independent coding method of depth images,while maintaining high virtual rendering performance in comparison.2. In order improve the quality of depth map coding and synthesis quality, adistortion correcting method for depth map intra coding is proposed. This method firstlyanalyses the phenomena of distortion pervasion in flat region and edge region,respectively. Then, the virtual view synthesis theory is introduced to analyse therelationship between distortion pervasion and synthesis error. Finally, a depth map intracoding method is proposed based on the pixel correction and distortion pervasioninhibition. Based on the distortion pervasion property of flat region and edge region ofdepth map, the blocks (or sub-blocks) which can cause the distortion pervasion arecorrected by the correction value. The huge number of following blocks (or sub-blocks)which were influenced by the distortion pervasion can obtain a smaller value of the ratedistortion optimization function. According to the experiment results, the proposedmethod can improves the quality of depth map and synthesis virtual view of3D video.3. A novel joint bit allocation method for the texture videos and depth maps codingis proposed to minimize the warping error, as well as improving the synthesis quality of 3d video (3DV). The contribution of the proposed method is that the Lagrangemultiplier is introduced in the bit allocation as a parameter to further reduce the virtualsynthesis distortion under the constrained bit. Firstly, the distortion model and bit modelfor texture videos and depth maps are constructed via the statistical experiments, wherethe Lagrange multiplier is treated as a parameter in these two models. Then, the modelof warping error is deduced by the comprehensive analysis on relationship between thecoding distortion and warping error. Finally, combing the above model, an objectivefunction for minimizing the warping error is designed and the unconstrainedoptimization method to solve the optimal Lagrange multiplier of the texture video anddepth map coding, with which the bit is allocated. Experimental results have verifiedthat compared with the full search based bit allocation algorithm and fix bit algorithm,an average gain of0.25dB and0.68dB can be achieved by the proposed method underdifferent coding conditions, respectively. Meanwhile, the coding bit of proposed methodis more approach to the constraint bit, and the average inaccuracy is only0.45%.4. In order to improve the quality of the reconstructed video and virtual view inRGB domain and reduce the video aliasing, edge blurring, and artifact, which arecaused by de sampling of chrominance component, two down/up sampling method forchrominance component of texture video are proposed in this paper. In first method, afrequency domain based down-/up-sampling filter is proposed to replace the spatialdomain sampling. With this algorithm, more original information can be preserved andaliasing can be avoided in chrominance components. Simulation results show thataverage0.21dB coding gain can be achieved by using the proposed algorithm comparedwith MPEG-4filter. In the second method, it is verified that the distortion introduced bychrominance sampling and coding can be approximately treated as a wide sensestationary noise, which can be suppressed by the Wiener filter. Accordingly, in thispaper, a two-dimensional Wiener filter is designed for reducing the noise of thereconstructed chrominance components, which further improves the video quality in theRGB color space as finally perceived by the users. Experimental results show thatcompared with bi-linear and inter-color up-sampling, the proposed method achieves theaverage gains of0.42dB and0.30dB, respectively. Meanwhile, the reconstructed videoobtained by the proposed method is more close to the original one in terms of subjectivequality.