Research On H.264 Rate Control Algorithm

At present the video compression has been used in a large scale. As one of the latest standards, the H.264 has significant improvements in coding efficiency compared with previous video coding standards. Rate control is an important technique in the H.264 codec, which has been thoroughly studied in MPEG2, MPEG4, H.263 etc. But in the H.264 standard, the research on rate control is more difficult than in other standards, so it needs further study.In this dissertation, we firstly introduce the main features of H.264. Compared to previous video coding standards, H.264 utilizes some advanced methods, along with increasing complexity at the same time. This dissertation discuss the method of rate control, which can be regarded as the port between the source and channel. We introduce the Rate-Distortion theory which analyze the relationship of rate and distortion of video, and study the Rate-Distortion optimization based on this theory, in order to minimize distortion of video, subject to a constraint rate. The source models are studied at the same time.The most influential rate control scheme of H.264 up to now is G012 scheme proposed by Li Zhengguo, and its improved counterpart-H017 scheme. We analyze the G012 scheme, and introduce the HRD model used in H017. In order to improve rate control scheme currently used in H.264, this dissertation propose the MARD model to estimate the complexity of image, and a novel rate control scheme based on it. Our scheme is better than existing H017 scheme in terms of control precision, bit fluctuation, and average image quality.The rate control scheme used in TM5 model of MPEG-2 is very simple and efficient, though it has some weakness. In this dissertation, we analyse the rate control scheme in TM5 model, use the idea of virtual buffer, and propose a novel rate control scheme with low complexity. This scheme is efficient, and don't need to update the coefficients of MAD prediction model and R-D model. It could be used in cases which is sensitive to compute complexity.