一种基于模型的自适应阈值分割算法

为了减少穷举式阈值分割方法中的重复计算，提出了连通域树（CCtTree)的结构与构造算法.在进行新阈值下的分割与连通域标记时，根据原阈值分割标记后得到的结果，结合新出现的连通域，以合并的方式得到新阈值分割下的连通域来减少多余的计算过程.给出了在CCTree中利用树搜索算法进行模型匹配区域搜索的方法.实际的图像库实验表明，在保证同样的模型匹配区域检出效果的基础上，基于CCTree的方法在运行时间上明显优于ETL，并能迅速有效地筛除重叠区域，获得更好的匹配区域.利用CCTree方法可以准确而快速地获得基于模型匹配的阈值分割结果.

Model based adaptive thresholding algorithm

To reduce computation overhead in exhaustive thresholding methods,a structure and construction algorithm of connected component tree(CCTree) was proposed.Through the labelling of the newly emerged regions and their merging with the existing components available as the result of previous thresholdings,the thresholding and connected component labelling under a new threshold was constructed incrementally,and the computation overhead was avoided.A method for model-matching region detection through tree searching on CCTree was proposed.Experimental results on a real image data base show that the CCTree based algorithm is superior to the ETL based algorithm while achieving the same model matching region detection effect,that the CCTree based algorithm can effectively filter out overlapping regions and obtain better matching regions,and that the algorithm can achieve accurate and fast thresholding outcome based on model-matching.