基于最小生成树和局部阈值的彩色图像分割方法

针对分水岭图像分割算法对于彩色图像的过度分割问题,本文提出一种基于最小生成树和局部阈值的解决方法。该方法主要利用图论中的最小生成树,对分水岭算法产生的过度分割区域进行合并。与其它的基于最小生成树的方法不同,该方法只有当构造出一棵完整的最小生成树时,才能计算出一个局部阈值,该局部阈值可确定原构造最小生成树过程的终止条件,进而可分割出彩色图像中的两个区域。重复上述过程,可分割出原彩色图像中的所有区域。实验证明,本文方法能够对彩色图像进行准确的分割,并且分割结果能很好地符合人眼的感知。

A color image segmentation method based on Minimal Spanning Tree and local thresholds

In order to solve the over-segmentation problem of watershed algorithm in color image segmentation,a method based on Minimal Spanning Tree and local thresholds is proposed.Using the concept of Minimal Spanning Tree in Graph Theory, the proposed method merges the over-segmented regions produced by watershed algorithm.Unlike other methods based on Minimal Spanning Tree,the proposed method calculates a local threshold only when a complete Minimal Spanning Tree has been constructed.This local threshold is used to determine when to stop constructing Minimal Spanning Tree,and then two regions can be segmented from the original color image.The above process is repeated until all the regions have been segmented from the original image.Experiments show that the proposed method can accurately segment color images,and the segmentation results can well meet human perception.