Study On Rate Distortion Optimization Method Of Depth Map Coding For Three Dimensional Video

Compared with the traditional two dimensional video,the three dimensional video can not only provide the textural and color information of the real scene,but also provide the depth information to the audience.This will lead to a more favorable immersive vision during the viewing process.Currently,three dimensional video has been paid a lot of attentions in these days and has been used in many areas,such as film industry,three dimensional game,medical image,and so on.As one of the efficient representations,multiview video plus depth?MVD?format stores the scene information as multiple texture videos and associated depth maps.Due to the good compatibility and the depth information,it has been chosen to be the standarded data format of three dimensional video by moving pictures experts group?MPEG?.However,the da-ta volume of color and depth videos is many times larger than traditional two dimensional video.Therefore,how to compress the MVD data efficiently becomes a hot issue.Depth maps,as one of the special digital image,have different characters from the texture videos.Depth maps are consist of a wide range of flat areas and sharp edges and contain the distance information between the object and the camera plane.Therefore,depth maps are not displayed to the audience but provide the geometric information to the view synthe-sis process.For this reason,the rate distortion optimization of depth map coding need to be based on the virtual view distortion instead of the depth distortion.By using the virtual view distortion in rate distortion optimization,the quality of virtual views can be improved efficiently.This dissertation focuses on the study of rate distortion optimization method of depth map coding.The main contributions of this dissertation are listed as follows.1.A fine virtual view distortion estimation method based on the change of disparity for mul-tiple pixels is proposed to calculate the virtual view distortion.The view synthesis process can be treated as a mapping process from the real view to the virtual view.The intensity of the real view is mapped to the corresponding position in the virtual view according to the associated depth map.The intensity at full pixel position in a virtual view is obtained by interpolation from the nearby mapping positions.When depth distortion occurs,the mapping positions will be changed.This will lead to an error of the intensity at full pixel position in the virtual view,namely the virtual view distortion.By introducing the change of disparity for the nearby mapping positions,the proposed method can calculate the vir-tual view distortion accurately.Experimental results demonstrate that the proposed method can improve 13.1%bitrate saving compared with the traditional depth distortion calculation method and can improve 1.2%bitrate saving compared with the virtual view distortion esti-mation method in 3D-HEVC reference software.2.This dissertation focuses on the Lagrangian Multiplier for depth map coding and proposes two algorithms to calculate the Multiplier.The first method is the adaptive Lagrangian Multiplier derivation model based on the quan-tization steps(Q_step)of texture videos and depth maps.The dissertation firstly analyzes the affect factors of Lagrangian Multiplier for depth map coding and finds out that the La-grangian Multiplier is not only influenced by the texture Q_stepbut also influenced by the ratio of depth Q_stepto texture Q_step.Then,the dissertation obtains the optimal Lagrangian Multipliers with different quantization parameters by the experimental method.At last,an adaptive Lagrangian Multiplier derivation model can be established through curve fitting by using these optimal Lagrangian Multipliers and the two affect factors.The proposed model can calculate the Lagrangian Multiplier for depth map coding accurately.Experimental re-sults prove that the proposed adaptive Lagrangian Multiplier derivation model can yield an average 3.13%bitrate saving compared with the traditional Lagrangian Multiplier method.The second method is a novel Lagrangian Multiplier calculation method based on the Rate-Quantization step(R-Q_step)model and Distortion-Quantization step(D-Q_step)model.In rate distortion optimization process,the Lagrangian Multiplier can be formulated as a func-tion of D and R.While,there exists a relationship among D,R and Q_step.Therefore,based on these observations,a novel Lagrangian Multiplier calculation method based on the Rate-Quantization step(R-Q_step)model and Distortion-Quantization step(D-Q_step)model is proposed in this dissertation.This algorithm represents the Lagrangian Multiplier for depth map coding as a function of texture and depth Q_stepand can calculate the Lagrangian Mul-tiplier accurately.Experimental results demonstrate that the proposed algorithm can lead to a more favourable coding performance for MVD coding.Compared with the traditional Lagrangian Multiplier method,the proposed algorithm can yield an average 3.42%bitrate saving.While compared with the proposed derivation model,the proposed algorithm can improve 0.28%bitrate saving.3.A rate distortion optimization model based on the reconstructed texture distortion for depth map coding is proposed in this dissertation.Since texture videos provide the textural and color information for the rending process of virtual views,the texture distortion will influence the virtual view distortion.On this basis,the dissertation firstly proves that besides the depth distortion and texture character,the virtual view distortion is also affected by the reconstructed texture distortion.Based on this analysis,by investigating the principle of virtual view distortion,a virtual view distortion estimation method is proposed in this dissertation.Moreover,the corresponding Lagrangian Multiplier can be calculated based on this distortion estimation method.At last,a rate distortion optimization model for depth map coding is achieved with a combination of the proposed virtual view distortion estimation method and the associated Lagrangian Multiplier.Experimental results demonstrate that compared with the traditional rate distortion optimization method,the proposed model can achieve an average 12.98%bitrate saving and an average 0.45 dB PSNR improvement.It can lead to a better coding performance for MVD coding.