基于模板聚类与综合的多模板快速定位算法

针对在一幅图像中定位多个模板的所有实例的情形，提出了一种基于多模板聚类和综合的快速目标定位方法。该方法首先使用带反馈的分级聚类算法对多模板进行聚类，并对每类模板用建立的数学模型综合出一个母板；然后，应用每类的母板在平移空间内搜索和匹配，且只在与母板相匹配的那些位置上才引导类内各子模板在该位置的匹配运算，最后用该算法对边缘图像进行了聚类、综合和匹配实验。实验结果表明，该算法在集成电路显微图像的多模板定位中是非常有效的。

A Fast Algorithm for Multiple Templates Locating Based on Templates Clustering

Aimed to locating all the instances of multiple templates in one image, a fast and more effective multiple templates locating algorithms based on clustering and synthesizing of templates is proposed. This algorithm can process those multiple templates even if only some of which are similar to each other. But all the templates must be almost the same size. First, a hierarchical clustering algorithm with feedback is applied to cluster the templates into some categories. In each category a mathematical model is applied to synthesize the templates in it. And thus a mother template is constructed. Second, the mother template of each category is used to search and matching in the translation space. And then the matched mother template is guided to check all the son templates. Edge maps are extracted for clustering, synthesizing, and matching. The partial Hausdorff distance matching with fast algorithm is suggested for mother template searching and matching procedure. Our algorithm is tested with difference multiple templates in integrated circuit micro images database. The results show that the scheme is efficient and effective for the task of multiple templates matching and locating.