Research On Rate Control For HEVC

It is reported that multimedia streaming over IP networks has been the dominant Internet traffic, thanks to the development of computer network, next generation communication technology and multimedia tools. Rate control plays a crucial role in video coding. As the communication bandwidth may be limited, an effective rate control algorithm should guarantee the visual quality of compressed video while make fully use of the bandwidth. High efficiency video coding standard(HEVC) is the newest standard for video compression. Previous rate control methods can hardly be used to HEVC without any modification because several advanced coding tools are adopted by HEVC. This dissert aims to develop effective and efficient rate control algorithms for HEVC.In this dissert, two rate control methods, which are based on an URQ model and a R-?ambda model respectively, are firstly introduced. Experiments are carried out to compare and analysis the performance of the two methods. Through analysis, it is found that the R-?ambda model based rate control method perform better than the URQ model based rate control method. But several drawbacks still exist in the model parameter updating and target bit allocation algorithm of the R-?ambda model. Firstly, the R-?ambda model takes only temporal correlation into account when updating the model parameters, resulting in inaccurate computing of Lagrange multiplier and quantization parameter(QP). Meanwhile, they do not consider the motion complexity of a largest coding unit(LCU) when allocating bits for it. Therefore, large mismatch may exist between allocated bit and real encoded bit. In this paper, we propose an effective LCU layer rate control scheme for HEVC exploiting both spatio-temporal context relationship and motion complexity.The new proposed method takes advantage of spatio-temporal context information to boost the performance. On the one hand, the real used Lagrange multiplier and generated bits of the high correlative coded LCUs in spatio-temporal context are carefully weighted to update the model parameters. On the other hand, the allocated bits for a LCU are adjusted based on its motion complexity to make sure that enough bits are assigned to the LCUs with high complexity. Thus, the improved Lagrange multiplier is more accurate to minimize the difference between the target bits and real generated bits. Extensive experiments are designed and carried out for evaluation. Experimental results show that the new proposed method outperforms the URQ method and R-?ambda model based method in rate control accuracy, visual quality, fluctuation of visual quality and rate-distortion metric, with negligible complexity increment.