An Enhanced Rate Control Initialization Method for H.264 based on Image Quality Balance of GOP Using Support Vector Machines

In this paper, we propose an enhanced method of initializing the quantization parameter (QP) in H.264 rate control algorithm based on image quality balance of GOP (Group of Pictures) using support vector machines (SVMs). Primarily, we analyze characteristics of some typical video sequences by motion vectors. And then we choose four video sequences with different characteristics and computer their spatial and temporal complexities and the optimal initial QPs which can guarantee to generate video sequences with consistent quality by minimizing the variation of QPs in a GOP, while ensuring the actual encoding bit rate closer to the target bit rate in various bit rates by proposed selection method of optimal initial QP. We utilize SVMs that can detect the optimal parameters by genetic algorithm to train the regression function of the extracted features that are the spatial and temporal complexities as well as given the target bit rates of sample video sequences and target that is optimal initial QPs. When any video sequence is given under any target bit rate, its spatial complexity is calculated and mapped to one of four samples through the proposed mapping method. Finally, its optimal initial QP is determined by regression function according to the mapped spatial complexity, corresponding temporal complexity of sample video sequence and given target bit rate.