基于层次聚类与峰值检测的FCM图像分割算法

摘要

针对传统FCM算法在图像分割中存在的过度依赖初始聚类中心、计算复杂度高等问题,结合层次聚类与直方图峰值检测,提出了一种新的FCM图像分割算法.首先根据图像灰度直方图统计信息 对图像进行层次聚类,然后将得到的聚类中心作为FCM算法的初始聚类中心对图像进行分割.该算法无需预先设置聚类数目,能自动搜索全局最佳聚类中心.实验结果表明,相比传统FCM算法和峰值检测的 FCM算法,该算法不仅可以有效地提高图像的分割效率,而且分割结果更加精确.%Aimed at the problem that the traditional FCM algorithm for image segmentation is excessively dependent on the initial clustering center and its computation is complicated, a new algorithm for FCM image segmentation is presented with hierarchical clustering and histogram peak detection.First of all, the statistic information on image grey scale histogram is taken as basis to conduct the hierarchical clustering of the image, and then the obtained clustering center is taken as initial clustering center of FCM algorithm to conduct the segmentation of the image.It is unnecessary for this algorithm to preset the number of clustering and the global optional clustering center can be automatically searched out.It is shown by experimental result that, compared to the traditional FCM algorithm, this algorithm will be able to improve not only the image segmentation efficiency but the segmentation result will also be more accurate.