Study Of Rate Control Algorithm Based On H.264/AVC

Rate control plays an important role in high performance encoder and runs through the whole process. As one of key technique in video coding, it controls the compression ratio. In all video coding standards, the performance of rate control will influences the reconstructed video quality and transmission ability directly. So a good rate control algorithm can get a better video coding performance.Nowadays, most researches of rate control algorithm about H.264/AVC are based on R-Q model and MAD prediction. The rate control algorithm based on sobel edge operator is one of good algorithm. In this paper we study this algorithm thoroughly. It can be found in experimental result that the advantage of this algorithm is current block's information are full used, and so the model parameter can be predicted accurately. But its computational cost is too high, and meanwhile the fluctuation of output bit rate is not smooth enough. So this algorithm can not be used in some real time and high complex channel circumstances. A rate control algorithm based on SATDc(Q)-R model is also introduced in this paper. This method uses current frame's pixel information fully, so it can get a good performance and a smooth output bit stream. But computational cost of this method is relatively high, so it also can not be used in some real time circumstances.After analyzing above method, two improved methods are proposed:Firstly, current block's complexity can be decided by a ratio that current block's SATDc(Q) is divided by the sum of surplus blocks' SATDc(Q) value. Then current block's target bit is computed according to this complexity. In this method, current block's pixel information is used fully. The improved rate control algorithm can get better performance and much smoother bit fluctuation, so it can be used in some more complex channel's circumstances.In addition, after analyzing the rate control algorithm based on SATDc(Q)-R model carefully, we found the reason of its high computational cost is that it needs many times quantization and dequantization. To solve this problem, a fast quantization method is proposed. Compared to ordinary quantization method, 80% or more computational cost can be saved, especially in low target bit rate circumstances.