Study On Rate Distortion Characteristics And Frame Coding Complexity In Video Coding

Rate distortion theory mainly computes the necessary bit rate to retrieval information source under limited distortion. If the output bits of an information source exceed in the network bandwidth, the information source need to be compressed, while the distortion introduced by the compression should not exceed the predefined limitation. Since the information of digital video is huge, how to sufficiently store these digital videos under limited bandwidth is an important issue in video applications, so video compression technology is a major and useful approach.As the theoretical basis of rate distortion optimization (RDO) in video coding, rate distortion (R-D) theory is widely studied. This thesis analyzes rate distortion theory and the modeling methods of rate distortion models, significantly analyzes model error of the quadratic model in H.264/AVC, and proposes the novel joint R-D model. According to the effect of the frame coding complexity in the rate distortion model and the influence of that in rate control performance, we propose novel models for intra and inter frame coding complexity respectively. Corresponding rate control scheme are introduced. Extensive experiments demonstrate that the proposed rate control algorithm can efficiently control the virtual buffer, precisely achieves target bit rate, obviously reduce unexpected frame-skipping, improve and smooth visual quality. Our researches include the following parts:1. Analysis of rate distortion characteristics and research on R-D model in video coding. This thesis significantly analyzes and verifies errors of the quadratic model, including the model error of the quadratic model, the error of simple MAD proportional model and the error of linear tracking model of JVT-G012. Furthermore, the exponential R-Q model is employed to analyze the joint R-D model.2. Proposes two models for intra and inter frame coding complexity separately according to different frame characteristics of two kinds of frame. Since intra encoding adopts prediction within a frame, the coding complexity is directly statistic by geometric gradient per pixel. For inter frames, five prediction methods are used to estimate the current frame coding complexity based on the previous coded frames, then the optimal coding complexity is selected to encode the current frame according to the coding performance. The experimental results indicate that these two models can simulate the actual frame coding complexity efficiently.3. Based on the frame coding complexity R-C models and the exponential R-Q model, we propose a novel joint R-D model, and introduce its new characteristics. Since the exponential model has derivation invariability, we simplify the model parameters and obtain its update method based on the first-order total differential of the joint R-D model. The 3-D image of the proposed joint model further demonstrates the linear relationship of R-C and the exponential relationship of R-Q.4. I- frame encoding has less computation and latency than P- frame encoding. Based on the advantages, we propose an intra-only rate control algorithm, which is implemented in reference software JM13.2 in H.264/AVC standard. The performance of the proposed algorithm is compared with that of JVT-W042, which is the first scheme proposed intra-only encoding. The simulation results demonstrate that our scheme outperforms JVT-W042. The overall performance of the proposed algorithm is excellent. And it achieves precise buffer control and better visual quality.5. Based on the optimal selection of frame coding complexity and the joint R-D model for P-frame, a rate control algorithm is proposed. This thesis also gives the parameter and encoder configures of the simulation for the proposed algorithms. The simulation results demonstrate that our scheme for P-frame achieves better visual quality than JVT does.