基于BIC准则的图像分割算法

A novel method for image segmentation based on the BIC

为解决图像分割算法中聚类数必须事先给定,否则无法进行聚类的问题,基于BIC准则建立了图像聚类算法的新型目标函数,提出了一个自动聚类算法.该算法实现了自动聚类,通过求聚类目标函数的最小值,给出聚类数,从而为聚类数的确定提供理论依据.对两幅图像进行实验模拟,实验结果表明,本文算法是有效的,具有普适性.当聚类图像灰度变换比较明显时,本文算法与K-均值、FCM算法聚类效果相同.当聚类图像灰度变换不明显时,本文算法的聚类效果优于K-均值、FCM算法的聚类效果.

Clustering algorithms need be given beforehand the classification number in classification,otherwise,classification can not be operated.This paper proposed an automatic clustering algorithm based on BIC by introducing a new type objective function.The algorithm not only realizes an automatic segmentation,but also provides theory bases for determining classification number which is obtained by solving the minimum value of clustering objective function.The simulation results show that the new method is valid and universal.When the transformation of grey levels of clustering image is obvious,the new algorithm,K-mean and FCM algorithm have the same clustering results.Otherwise,when the transformation of grey levels of clustering image is not obvious,clustering results of the new algorithm is superior to K-mean and FCM algorithm.