一种基于硬件实现的快速运动估计半像素级搜索算法

A fast half-pixel motion estimation algorithm based on hardware implementation

研究了在绝对差和准则下的整像素级块匹配和半像素级块匹配的联系,对绝对差和进行了合理的数学曲线拟合,通过数学曲线来预测最小绝对差和所在半像素位置,从而得到半像素级最佳匹配矢量。分析了三种不同的凹函数预测模型,提出了一种适合硬件实现的运动估计快速半像素级搜索算法。该算法直接根据整像素级运动估计的结果来推算半像素级运动估计结果,在很大程度上降低了半像素级运动估计的运算复杂度,从而利用低码率视频编码的实时实现。试验结果表明该算法可获得较好的重建图像质量。该算法利于硬件实现,可以方便地集成到现有的视频编码器中,具有较好的实用价值。

The internal relation between full-pixel block matching and half-pixel block matching under the block distortion measure of sum of absolute difference (SAD) is studied. The SAD is modeled by mathematical curves, through which the half-pixel position that has the minimal SAD is predicted and the best half-pixel accuracy motion vector is calculated. A fast half-pixel motion estimation algorithm is proposed after analyzing three different mathematical prediction models based on concave functions. The result of half-pixel motion estimation is directly deduced from that of full-pixel motion estimation, which greatly reduces the computational complexity of half-pixel motion estimation and eases the implementation of low bit-rate real-time video encoder. Experimental results show that this algorithm can keep good quality of reconstructed images as well. The algorithm can be easily integrated into video encoders available.