摘要
为了得到FCM聚类多阈值分割中最佳聚类个数,针对Bezdek熵在数字图像数据聚类有效性判别中的不足,提出一种改进的聚类有效性判别函数.新函数通过在Bezdek划分熵中增加补偿项来突出最佳聚类时的函数值,提高有效性判别的正确性.试验结果表明,基于改进初始隶属度矩阵生成方法的FCM算法,计算迭代次数为传统FCM方法的55%,计算用时减少了约45%,而且由改进聚类有效性判别函数得到的最佳聚类数目和试验图像相符,效果明显优于Bezdek熵方法,由最佳聚类数得到的分割图像能够体现目标绝大多数信息,证明了本算法的有效性和正确性.
In order to calculate out the optimal class number of FCM cluster algorithm,an improved cluster validity function was proposed at the basis of Bezdek partition entropy.A compatriotic term was added to Bezdek partition entropy for the purpose of making minimum function value stood out and bettering the correctness of calculating result.Experiments was performed and the results shown that the number of iteration based on the new initial partition matrix creating method was decreased about 55% and the searching time was reduced by 45% compared with traditional method,the results also shown that the optimal class number obtained from our function was agree with tested image and was much better than results obtained from Bezdek algorithm,the experiments also told that the segmentation image with optimal class number embodied most of object information of initial image.
出处
《安徽大学学报（自然科学版）》
CAS
北大核心
2007年第5期23-27,共5页
Journal of Anhui University(Natural Science Edition)
关键词
聚类有效性
FCM
最佳聚类数
阈值分割
隶属度矩阵
cluster validity
FCM
optimal class number
threshold segmentation
partition matrix
