基于中心点预测的分数像素运动估计改进算法

Improved Fractional Pixel Motion Estimation based on Center Prediction

针对H.264采用的分数像素运动估计算法存在计算复杂度高的不足,通过探索分数像素点之间相关性强及分数像素搜索范围为单峰值误差曲面的特点,提出了一种基于中心点预测的分数像素运动估计快速算法.测试结果表明:与全搜索分数像素搜索相比,该方法平均减少分数像素搜索点数69.32%,峰值信噪比相同,码率增加4.48%.该算法在保持图像质量不变和码率增加较少的情况下能大大减少搜索点数目,有效提高了分数像素运动估计的速度.

Fractional pixel motion estimation has been applied in H. 264 to achieve higher compression. It, however,distinctly increases complexity in the procedure of motion estimation. The higher eorrelation and the unimodal error surface inside fractional pixel search window were explored in this paper, and thus a fast fractional pixel motion estimation algorithm based on center prediction was proposed. Simulation results show that the proposed algorithm maintains an unnoticeable＇quality loss in PSNR and reduces fractional pixel search points by about 69. 32 %, while increases bit rate by only 4. 48 % when eomparing to full search strategy. The proposed algorithm is efficient to speed up the fractional pixe[ motion estimation with preserving the image quality and increasing bit rate little.