改进的KFCM算法及其在感兴趣区域提取中的应用  

Improved KFCM Algorithm and Applied in Extracting Interesting Areas

作　　者：辛晚霞 杨明[1] 

机构地区：[1]中北大学,山西太原030051

出　　处：《电视技术》2014年第11期35-37,41,共4页Video Engineering

基　　金：国家自然科学基金项目(61171179)

摘　　要：基于核函数的FCM算法(KFCM)是一种常用的聚类算法,它需要人为地确定分类数,对噪声比较敏感。针对感兴趣区域提取问题,提出一种改进的KFCM算法,该算法先用k'-means算法估计分类数,再用KFCM算法进行聚类,改进隶属度函数,使新算法的隶属度为其邻域隶属度的平均值,提高了算法的抗噪能力。将新算法应用到感兴趣区域提取中,实验结果表明,新算法不需要人为地确定分类数,并且相比传统的FCM算法和KFCM算法能更有效地抑制噪声。Kernel-based FCM algorithm（KFCM） is commonly used in clustering,it needs to artificially determine the number of categories,and is sen- sitive to noise. An improved KFCM algorithm is proposed to extract interesting areas,the algorithm first estimates the number of categories by using k＇- means algorithm ,and then adopts KFCM algorithm to cluster. In order to improve membership function and enhance noise immunity ,the new algorithm lets membership function be the average of neighbor membership function. The new algorithm can be applied to extract interesting areas,the experiment shows that the new algorithm does not need to artificially determine the number of categories,and it is more effective than traditional FCM algorithm and KFCM algorithm in suppressing noise.

关 键 词：FCM算法 KFCM算法 k'-means算法 图像分割 感兴趣区域提取 

分 类 号：TN911.73[电子电信—通信与信息系统]