Chinese Journal of Computers   Full Text
Title	Efficient Rate Control Schemes for H.264/AVC
Authors	YUAN Wu1) LIN Shou-Xun2) NIU Zhen-Dong1) LUO Hai-Yong2) ZHANG Yong-Dong2)
Address	1)(School of Computer Science, Beijing Institute of Technology, Beijing 100081) 2)(Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100080)
Year	2008
Issue	No.2(329—339)
Abstract &   Background	Abstract In this work, a novel coding characteristics prediction scheme is presented to improve R-D modeling, by exploiting spatio-temporal correlations. Two different approaches to the problem of optimum bit allocation at a macroblock-by-macroblock basis are achieved, one of which is developed on a modified MPEG-4 Q2 rate model and the other on a linear rate model. Extensive experiments show that the linear scheme is a bit more accurate than the quadratic one while they achieve similar coding performance. It’s also shown that both the two schemes significantly exceed JVT G012, the current standardized RC scheme. keywords rate control; optimum bit allocation; H.264/AVC background Rate control (RC) is one of the key problems in regard of coding performance, and has drawn significant research attention. For rate control, its main task is to regulate DCT coefficients quantization to adapt to the actual channel bandwidth, and meanwhile achieve high-quality reconstructed pictures at the decoder. Several RC schemes have been reported in the literature, including TMN-8, TM-5 and VM-8 etc. Generally speaking, a typical RC scheme can be decomposed into two steps, which are bit allocation and subsequent achievement of the bits target. The insight behind optimal bit allocation and accurate achievement is to precisely approximate rate-distortion(R-D) behaviors of video content via mathematical modeling. Because video coding is a highly nonlinear process, it is a difficult task to precisely approximate R-D behaviors by using a close-form formula, and an empirical approach is most often used indeed. Based on R-D models, Lagrangian optimization or linear programming methods are employed to achieve optimum bit allocation. As a new generation of video coding standards, H.264/AVC greatly outperforms prior coding standards in coding performance by exploiting lots of complicated coding methods. As a side-effect, H.264/AVC gets highly complicated. The introduction of Lagrangian coder control method into a H.264/AVC-complainted coder greatly challenges the design of RC schemes. The Lagrangian coder control method demands quantization parameter as a prerequisite to the RDO for inter/intra prediction, and consequently couples inter/intra prediction and RC much tightly. Since accurate R-D modeling on the residual can only be conducted after inter/intra prediction, the peculiarity of Lagrangian coder control method leads to the chicken and egg dilemma. Thus, the design of RC scheme for H.264/AVC is quite different from prior standards, and conventional RC scheme cannot be directly employed in a H.264/AVC-complianted coder. Several works related to RC have contributed to H.264/AVC, such as JVT-F086, JVT-G012 and JVT-O016. In JVT-F086, a multiple-pass scheme is presented to circumvent the dilemma, while in JVT-G012 and JVT-O016, a linear MAD model is used indeed. In JVT-G012, the conventional MPEG-4 Q2 model is employed to calculate the quantization parameter. Compared with JVT-G012, JVT-O016 improves the accuracy of MPEG-4 Q2 and MAD models by using an effective coding characteristics prediction scheme. Moreover, JVT-O016 achieves optimum bit allocation at a macroblock-by-macroblock basis in a TMN-8-alike way, and therefore significantly improves the coding performance. In this paper, an approach similar to JVT-O016 is presented with significant difference in using a linear rate model instead of a quadratic rate model and a quadratic distortion model instead of a linear distortion model. Based on the R-D models, we develop a close-form formula to the problem of optimum bit allocation.