Research On Rate Control Algorithm For3D Video Coding

Three Dimensional (3D) video has become one of research highlights in the academic andindustry circles, and will have extensive application prospects. However, because of a large amountof3D video data, the compression coding and transmission of3D video data become very difficult.Rate control is one of the key technologies for the3D video transmission. Compared with TwoDimensional (2D) video, the coding prediction structure, the frame types, the relationship modelbetween rate and quantization parameter, and the quality fluctuation of each view need to be morein-depth consideration. So the complexity of rate control algorithm is greatly enhanced. Then,owing to the diversity of the real world, there exist various scene representations to build3D video.We need to specifically design the appropriate rate control algorithm. Finally, JMVC (JointMultiview Video Coding), which is multiview video reference software, has not yet given the codeof rate control. This paper mainly focuses on rate control algorithms for3D video coding based onH.264/AVC. And the contributions are:(1)For the stereoscopic video representation, we propose a novel macroblock level ratecontrol scheme based on binocular just noticeable difference (BJND) model for stereoscopic videoin this paper. The proposed rate control algorithm is performed on four layers, namely view layerbit allocation, group-of-pictures (GOP) layer rate control, frame layer bit allocation andmacroblock level rate control. In the view layer, different proportions of bit rate for the left viewand right view are allocated by pre-encoding the first GOP. In the GOP layer, the target bit rate isallocated for each GOP, and the initial quantization parameter is determined for each GOP as well.In the frame layer, the target bits allocated to each frame are computed. In the macroblock layer,visual weighting factor, which is measured by the BJND value of macroblock, is used to adjust themacroblock level bit allocation, so that the rate control results in line with the human visualcharacteristics. Experimental results show that compared with MVC rate control algorithm ofZheng and SMBRC, the proposed scheme can get smaller rate control error and better subjectivequality of video images by image detail contrast.(2)Multiview video plus depth (MVD) representation can efficiently describe the actual3Dinformation, while providing the backward compatibility with the existent2D video. Therefore, itis significant to design rate control scheme for MVD. This paper presents a novel rate controlscheme for multiview video and depth coding system. The rate control algorithm is performed onfour layers, in the video and depth layer, the optimal bit allocation ratio between video and depth isselected by statistical experiments. And the optimal quality of virtual view rendering is got underthe less total rate. In the view layer, different proportions of bit rate for each view in video or depthare allocated by pre-encoding the video or depth. In the GOP layer, the target bit rate is allocatedfor each GOP, and the initial quantization parameter is determined for each GOP as well. In the frame layer, the frame complexity is calculated by encoding quantization parameter multiplied bybits. Then the frame layer target bits are allocated by the relative complexity of current frame.Experimental results show that the proposed scheme can control the bit-rate accurately to meet therequirements of3D video systems. And compared with the Fixed Video/Depth4:1in Liu’salgorithm, the proposed algorithm has better rate-distortion performance.(3)The joint coding of texture videos and depth maps plays a practical importance in3D videosystem. Besides coding efficiency, the strategy of bit allocation between texture videos and depthmaps is key issue. The paper firstly analyses the impacts on the quality of virtual views from thejoint coding of texture videos and depth maps. Then the paper explores the relationship andestablishes a quadratic model between texture video quantization step Qtand depth mapquantization step Qd. Consequently, the paper proposes a joint bit allocation scheme based on themodel to get the optimal quality of the synthesized views. Experimental results show that theproposed scheme has a better rate-distortion performance for the synthesized views, compared withfixed bits ratio between texture videos and depth maps, i. e.5:1. Moreover, compared with the jointbit allocation algorithms of Morvan, Liu and Yuan, the proposed algorithm has very lowcomplexity.