基于下采样的快速块匹配搜索算法及降噪应用

提出一种基于下采样的快速块匹配搜索算法——下采样三步搜索算法(DTSS)。在视频序列中，通过双线性插值法下采样得到当前帧和参考帧的采样帧，在采样帧中进行基于块的三步法初步运动估计，然后通过下采样恢复进行精细搜索，得到当前帧的运动矢量场。利用双线性插值下采样方法的低通特性，可以实现在噪声干扰情况下对运动矢量的准确搜索；另一方面，采用下采样技术，使得块匹配搜索算法的搜索速度加快，达到快速搜索目的。实验结果表明，在保持搜索准确度和提高搜索速度方面，DTSS明显优于三步法和菱形搜索等经典的块匹配搜索算法。最后结合经典的多假设运动补偿滤波(MHMCF)算法验证了DTSS应用于视频图像降噪中的有效性。

Fast block-matching search algorithm based ondown-sampling and its application in denoising

A fast block-matching search algorithm based on down-sampling named Down-sampling Three Step Search (DTSS) was proposed. In video sequence, the down-sampling frames of the current frame and the reference frame would be obtained first by bilinear interpolation sampling. Then, the initial motion estimation would be estimated by using Three Step Search (TSS) in the sampling frames. At last, refined search results would be obtained according to the relation between sampling frames and original frames. Due to low-pass characteristic of bilinear interpolation, DTSS can effectively suppress noise and get accurate search of motion vector. On the other hand, while DTSS bears the advantage of sampling technique, it can make block-matching search algorithm faster. The experimental results show that DTSS algorithm outperforms the present searching methods like TSS and Diamond Search (DS) both in search speed and accuracy. Combining this algorithm with the Multi-Hypothesis Motion Compensated Filter (MHMCF) noise reduction algorithm, the experimental results also indicate that this block based search algorithm suits well in video denoising applications.