Research On Rate-distortion Optimization Technology For Single-view And Multi-view Video Coding

H.264/AVC and its extension IV are the newest international single-view video coding and Multi-view video coding (MVC) standards, which are jointly developed by the ITU-T VCEG and the ISO/IEC MPEG standards committees. H.264/AVC has been concerned widely because of its high coding efficiency and network friendliness, MVC has also attracted a lot of researchers to study it. These two areas are the most active in the research field of single-view video coding and MVC, to which multimedia industry and many research institutions have also paid much attention. So it is significant for researches to have an in-depth research on the key techniques used in video coding. As rate distortion optimization (RDO) is not limited by coding framework, it is widely used in video coding, such as motion estimation, mode selection, rate control, error control etc. RDO is also one of the most important ways to improve efficiency. In this dissertation, the application of RDO in single-view video coding and MVC are studied. The content of this dissertation includes the following 4 aspects:First, The fast RDO algorithm for single-view video coding are studied. The similarities of bits, coded block pattern (CBP) of a coding macroblock(MB) between each inter mode and optimal mode are investigated. The results show that the combination of bit rate and CBP can distinguish coding mode accurately. Based on these characteristics, a novel fast RDO algorithm combing with certain MB characteristics is proposed. This algorithm employs the P16×16 mode's reference frame to select the candidate reference frames for other inter modes. Bits and CBP that select the candidate modes are the coding results of mode 16×16. The candidate modes are selected by bits and CBP which are the coding results of mode P16×16. The optimal mode will be selected in all candidate reference frames and modes with RDO. The algorithm is evaluated by using a wide range of test sequences of different classes. Experiment results show that the proposed algorithm has overcome the drawback existing in many algorithms that the saving of the coding time is depended on, more or less, the temporal and spatial correlation. Compared with RDO in JM12.0, the experiments results demonstrate a reduction coding time by 71.20% on average, with 1.32% of bits increase and 0.0752dB of Luma PSNR loss.In single-view video coding framework, the error propagation characteristics of H.264/AVC video streaming transmitting in error-prone channel is analyzed in detail. Based on the existing model, which are used for estimating the transmission distortion by channel simulation, a precise simulative estimate of channel distortion model is put forward. In this model, specific packet loss-rate channel distortion is calculated by a few specially-designed channel simulators which according to the error pattern to simulate video transmission at encoder firstly. Then, the relationship between packet loss-rate and channel distortion is studied. The relationship between them is then used to calculate other packet loss-rate channel distortion. The error pattern designed in this model overcomes the defect of uneven MB lost, and the quadratic relationship between packet loss-rate and channel distortion decreased the estimation errors of other packet loss-rate. There is no distortion accumulation and no need to decode picture at encoder during the estimation process. The experiments results show the average estimation error is 9.4% with encoding time increased 10%.Based on the MVC framework proposed by HHI, the estimation model of transmission distortion is investigated. Firstly, the framework's structural characteristics are analyzed in detail. Then, the advantages and disadvantages of recursive optimal per-pixel estimate model and channel simulative estimate model on MVC transmission distortion estimation are carefully compared. According to the results, a hybrid estimation model is proposed, which combines the advantages of recursive optimal per-pixel estimate model with channel simulative estimate model. The transmission distortions of key frames and non-key frames can be estimated by channel simulative estimate model and recursive optimal per-pixel estimate model respectively. The estimation accuracy is demonstrated via simulation results for video transmission over packet loss channel. With a little encode time increase, the average estimation error is 15.91% for different intensity sequences at different packet loss rate. It has illustrated that the transmission distortions can be estimated accurately and quickly by proposed model.Finally, a joint source-channel RDO mode selection algorithm in MVC is proposed. In this part, the characteristics of transmission distortion and the importance of protecting key frame are discussed. It is necessary to select different error control methods to encode key frame and non-key frame. Based on the analysis, the proposed model combines the transmission distortion estimation model with traditional RDO to select optimal mode with the purpose of decreasing the distortion of decoded pictures. To protect key frames, smaller Lagrange multiplier is adopted in key frame coding, and this Lagrange multiplier is decided by a large number of statistical experiments. Experiments results show that the average PSNR increased 1.5 dB with bit increased 13%. These results indicate that adopting different Lagrange multipliers in key frame and non-key frame is a good way to guarantee the quality of the decode picture, moreover, the bit increase can be controlled in a reasonable range.