为了更好地选取图像阈值,将最大模糊熵(MFE)准则与最大互信息(MMI)准则结合,提出最大模糊互信息(MFMI)准则。同时为了有效确定最佳分割类数,提出根据模糊互信息差(dFMI)来判别的准则。综合上述的两点改进,提出一种新的多阈值分割算法——最大模糊互信息量分割算法(MFMI)。对合成图像、无损检测图像、标准测试图像进行仿真,同时对比结合前的MFE与MMI,经典的阈值分割法如OTSU和MET,以及流行的模糊C均值算法(FCM),可以发现MFMI误判率最小,代价是运行时间较长。综上,MFMI是一个有效的图像分割方法。
A novel multi-threshold image segmentation method,Maximum Fuzzy Mutual Information(MFMI),is proposed on the basis of MFMI principle which combines the Maximum Fuzzy Entropy(MFE) and Maximum Mutual Information(MMI) principles to adopt the threshold more efficiently,and differential Fuzzy Mutual Information(dFMI) principle to determine the best number of clusters.Experiments on synthesized images,nondestructive testing images,and standard testing images demonstrate that MFMI is the most powerful in terms of classification error and penultimate in respect of computation time compared with MFE and MMI,OTSU and MET,FCM.In conclusion,MFMI is a valid and effective method for image segmentation.