采用下采样和位平面匹配的局部运动估计

Digital image stabilization based on sub-sampling and bit plane matching

针对电子稳像系统的实时性要求,提出了一种基于下采样、位平面匹配和三步搜索策略的快速局部运动估计算法。该算法将搜索比较次数由 225 次减少为 24 次而精度基本不变;利用图像下采样技术将参与估计的子图像像素增加 4 倍而运算量基本不变,提高了由于子图像中运动物体引起干扰的抵抗能力。实验表明,该算法搜索速度明显提高,而精度基本不变,所得到的匹配位置非常接近全搜索匹配法获得的匹配位置。

In view of the timeliness demands of an Digital Image Stabilization (DIS) system, a fast local motion estimation method is proposed, which combines the advantages of the sub-sampling technology, bit-plane matching (BPM) and the 3 step search (3SS). The algorithm cuts the search and comparison times from 225 to 24 while keeping its accuracy invariable. The pixels of the sub-image participating in the estimation have been increased by a factor of 4 by based on sub-sampling technique but its operation quantity keeps invariant. This improves the capability of anti-interference induced by the movement of sub-image object. The experiments show that the search speed of the algorithm has been obviously improved but its accuracy keeps invariant. The obtained matching position very much approaches that obtained by full search matching algorithm.