Research On Rate-Distortion Optimization And Bit Allocation Techniques For CGS/Spatial Scalable Video Coding

Rate-distortion optimization (RDO) technique is one of the key techniques in video coding, which is based on Shannon’s rate distortion coding theory. RDO is used to strike a balance between bit consumption and video distortion. This dissertation focuses on the RDO and the application of rate-distortion coding theory in scalable video coding. The main contributions of this dissertation are listed as follows:Firstly, Conventional rate models characterize the rate behaviors of residual texture only, as a result, the header bits cannot be calculated properly. Thus, the accuracy of the rate model needs to be questioned. In order to obtain an accurate rate model, the rate characterization of header information is considered by analyzing the relationship between the percentage of header bits and quantization parameters. The experimental results show that the accuracy of the rate model is significantly improved by analyzing characteristic of residual texture bits and header bits roundly.Secondly, the mean absolute difference (MAD) of the residual texture in the base layer (BL) is not always the suitable reference for inter-layer prediction. In order to obtain accurate MAD prediction values of the residual texture in the enhancement layer (EL), we develop a novel switched model which considers the MAD from previous temporal frames and previous spatial frames together. Based on the novel MAD prediction model, the R-D models of the BL can be extended to the EL flexibly. The experimental results show that the accuracy of the residual MAD prediction in the EL is significantly improved by the switched model. Finally, to avoid collecting all the R-D data while retaining optimality, an efficient bit allocation algorithm for H.264/SVC using the R-D models is devised. The bit allocation problem for H.264/SVC is formulated as an optimal quantization step size decision problem, which targets to minimize the total distortion under the target bit rate. Moreover, the Lagrange cost function is introduced to map the constrained optimization problem of the bit allocation to an equivalent unconstrained optimization problem. The performance of the proposed algorithm is conducted on various test sequences. Experimental results show that the proposed bit allocation algorithm outperforms two prior H.264/SVC bit allocation algorithms served as benchmarks.