Research On Rate-distortion Optimization Techniques Of Depth Coding For 3D-HEVC

Recently,the demand for Three Dimensional(3D)video is largely increasing because of its capability of providing real depth perception and immersive visual enjoyment for multimedia applications.Currently,the most widely used representation of 3D video data is Multi-view Video plus Depth(MVD)format,which enables a high quality representation of 3D world video content and facilitates the depth image based rendering(DIBR)technique to generate additional virtual texture views between original views.Meanwhile,the high data volume of MVD format puts forward a challenge for the coding and transmission of 3D video data.This paper focus on the research of efficient depth coding.The rate-distortion optimization(RDO)is used to select the coding mode.In the process of depth coding based on 3D-HEVC,it is an effective way to guarantee the quality of virtual view synthesis with taking synthesized view distortion change(SVDC)as distortion metric according to the characteristics of the depth map and its application scene.Based on the RDO technique of video coding,this paper discusses the effect of texture information on the depth coding,and explores the improvement of the coding rate-distortion performance of depth maps in 3D-HEVC.The computation of SVDC requires pixel-based rendering operation in the encoding process.The scheme can achieve better coding performance,but its computational complexity brings severe burden to the encoder.Therefore,the synthesized view distortion estimation(VSDE)model is proposed to improve the depth coding efficiency.With converting the definition of view synthesis distortion into the mathematical model of pixel level area distortion,the distortion estimation model based on the fine area is proposed in this paper,improving the performance of the existing view synthesis distortion estimation model.Compared with the original distortion estimation model,under the same virtual view quality,an average 1.03% bitrate saving can be achieved by the proposed algorithm.Furthermore,this paper analyzes the Lagrangian multiplier in the rate-distortion cost function.After the distortion metric is changed,it is necessary to adjust the Lagrangian multiplier to achieve a certain balance between the bit rate and the distortion,aiming to select the optimal coding mode.Since the synthesized view distortion is affected by the depth map istortion and the texture distortion,the fact that the Lagrangian multiplier in RDO of depth coding is related to not only the depth quantization parameter(QP)but also the corresponding texture QP is verified by experiments.On this account,the depth Lagrangian multiplier selection method based on texture distortion is proposed.Under the same synthesis view quality,the total bit rate of texture and depth is reduced by 1.36% in average.