一种基于Kalman滤波的4×4块运动估计方法

Kalman Filter for Motion Estimation Based On 4×4-block

为了更加准确的估计运动矢量,提高运动补偿性能,提出了一种基于Kalman滤波的4×4块运动估计方法。该方法通过宏块内相邻子块的运动相关性形成基于4×4块Kalman滤波器,根据Kalman滤波器多子块预测自修正来自三步搜索(TSA)的16×16宏块级运动矢量值,从而估算出最优宏块级运动矢量值,有效的提高的运动估计精度。试验表明,特别是在运动较为剧烈和局部场景瞬时变化时,该方法比TSA和16-KF提供了一个更大的平均PSNR。

In this paper ,a technique of Kalman filter based on 4×4-block algorithm is proposed to improve the motion estimates and enhance the motion compensation. According to the correlation of motion in intra-macroblock, we can utilize more sub-blocks motion estimates predicted form Kalman filter based on 4×4-block to amend the estimate of motion, which is obtained through the three step algorithm.In the sense that we can estimate optimum motion vectors based 16×16-block to have a better motion estimate .Experimental results have shown that, specially when there exists fast motions and instantaneous changes of the local scenes, this algorithm provides a greater average PSNR than TSA and 16-KF.