Research On Fast Video Coding And Rate Control Technologies For Multi-View Video Based On 3D-HEVC

Multiview video plus depth information is the most efficient representation format of three dimensional(3D) video, which supports reconstruction of an arbitrary virtual 3D video scene within a certain range and provides stereoscopic and immersive perception of real-world scene. With the overwhelming expanding of video applications, multiview video coding technology and its related applications have gained widely conceren by academia and industry. They have become one of the research interests in the field of multimedia information processing. However, multiview video involves huge amount of information. To satisfy the requirenments for effiecent transmission and processing of multiview video content, effective compression and transimmision schemes must be proposed. High Efficiency Video Coding(HEVC) Based 3D Video Coding(3D-HEVC) standard has been finished development and standardization by the Joint Collaborative Team on 3D Video Coding Extension Development(JCT-3V), with the goal to achieve higher compression performance and better network adaptation. It has been already been issued as the latest 3D video coding standard in February 2015.This thesis focuses on the investigation of multiview video coding and rate control technologies, especially on facilitating the adoption of the emerging international 3D video coding srandard 3D-HEVC, to speedup the multiview video coding process and to control the bitrate more accurately. The main work and contributions of the thesis are as follows:1. Basic principles and key coding techniques of the High Efficiency Video Coding(HEVC) are summarized. Multiview video coding and rate control techniques are presented for multiview extension of HEVC. Two prevalent multiview video coding frameworks are analysed and compared. In this thesis, the coding framework of 3D-HEVC is adopted due to its better coding performace.2. A fast inter-prediction texture mode decision method is proposed for multiview video coding to accelerate texture view coding process. By analysis of inter-view correlation of coding modes among neighboring views and hierarchical depth correlation of coding modes in the quadtree block partitioning structure, fast merge mode decision methods are separately designed for B frame and P frame to early determine merge mode coded CUs. Then, fast skip mode decision methods are proposed to further determine whether the current CU is a merge or skip mode coded CU. By the proposed method, unnecessary prediction mode decision process can be eliminated. Experimental results show that our algorithm can significantly reduce the complexity of texture views with negligible rate distortion performance loss.3. By design of optimized rate-distortion and coding parameter combined model correlating target bitrate and coding parameter, a largest coding unit(LCU) level rate control method is investigated to improve the rate control accuracy. An optimized bit allocation method and an improved model parameter predictor selection scheme are proposed for rate distortion model. In combination with the Mean Absolute Differences(MAD) from temporal and inter-view reference frames, reliable target bitrate and model parameter predictors can be achieved by adaptively assigning MADs of corresponding LCUs in reference frames to different prediction direction as prediction weights. By investigation of relationship among model parameter predictors, target bitrate and coding parameter lambda, a bi-directional prediction based optimized R-Lambda model is proposed for accurate rate control of texture views in multiview video. Experimental results show that the LCU level optimized R-Lambda model based rate control algorithm can achieve better rate distortion performance and smoother bitrate fluctuation while maintaining a similar bitrate accuracy compared with the original R-lambda model based rate control algorithm.