Research Of H.264Rate Control Algorithm

H.264is a new video coding standard and its purpose is to provide high video quality video in the condition of low bit rate transmission. In order to achieve the purpose, too much new coding techniques are applied to H.264.RDO is one of the technique, it can make video redundancy low. Although video performance has a great improvement, in order to maintain a stable video quality during transmission, rate control must to be considerate. In the standard of H.264, the mean absolute difference of the MB can be obtained after rate distortion optimization. As a result, traditional rate control algorithm is not suitable for the H.264coding standard.In order to solve the above problems, JVT-G012proposed a method of using a linear prediction algorithm to predict the MB’s MAD values of the basic unit layer. In the purpose of rate control, QP value can be calculated through MAD. But, there are still some problems in this algorithm, which includes inaccurate MAD value prediction. Coefficient of the MAD prediction scheme can be updated by linear regression, which can bring large amount of computation complexity. At the same time, the algorithm only considers the temporal correlation, which leads to the inaccurate prediction.A weighted spatial and temporal MAD prediction model is proposed to solve the above problem. The model takes into account not only the temporal correlation, but also considers the spatial correlation. In the new temporal correlation prediction, we use two MBs to calculate the temporal correlation MAD value. We also find that some video sequences have weak temporal correlation but strong spatial correlation. Four directional spatial prediction is proposed, including vertical, horizontal, upper-left, upper-right and we give the method to choose the suitable directional scheme. After we get the temporal MAD value and spatial MAD value, we can calculate the new weighted MAD value of the MB by the feature of the video sequence. The proposed weighted MAD prediction scheme is tested on the platform of JM. In order to make more accurate experimental data, we select some sequences of different features, not only including some simple sequences and complex sequences, but also some sequences with scene switching. The final experimental results show that the actual bit rate is much too close to the target bit rate.The bit stream is more stable. The PSNR also increases in a certain degree, which means the quality of the video encoding has a great improvement. In some sequences, the PSNR is up to0.68dB.That means the video coding quality is good.