为了解决原始最好兄弟相似性(BBS)算法在剧烈非刚体变形、部分遮挡以及非均匀光照等复杂环境下匹配失败的问题, 提出了一种更加稳健的模板匹配算法。将曼哈顿距离替代欧氏距离作为两个图像块之间的相似性度量, 在此基础上, 滑动窗口逐像素匹配得到新的由BBS响应值构成的置信度图, 对该置信度图进行阈值筛选, 并对剔除较小值后的置信度图滤波处理后, 将最亮连通区域的中心位置定位为匹配结果。实验与分析结果表明, 该算法可以有效地解决在弹性变形、相似区域干扰、部分遮挡与剧烈光照变化等变换与干扰存在情况下的图像匹配定位问题。

To solve the problem of image matching failure under complex conditions including extreme non-rigid transformation, partial occlusion, and imbalanced illumination, a more robust template matching algorithm than the original best-buddies similarity (BBS) algorithm is presented. The similarity measurement between image patches is represented by Manhattan distance instead of Euclidean distance. On this basis, the new confidence map is constructed by sliding window to compute the BBS response values. The center of the brightest connected region is determined to be the last matching location after threshold filtering the confidence map by filtering process to eliminate the minor response values. Experimental and analysis results show that the proposed algorithm may be used to match the images with elastic deformation, similar region interference, partial occlusion, and extreme illumination change, etc.