基于入侵性杂草克隆的图像聚类方法  

Image Clustering Method Based on Invasive Weed Colonization

作　　者：苏守宝[1] 方杰[1] 汪继文[1] 王本有[2] 

机构地区：[1]安徽大学计算智能与信号处理教育部重点实验室,安徽合肥230039 [2]皖西学院计算机科学与技术学系,安徽六安237012

出　　处：《华南理工大学学报（自然科学版）》2008年第5期95-100,105,共7页Journal of South China University of Technology(Natural Science Edition)

基　　金：国家“973”计划项目(2004CB318108);国家自然科学基金资助项目(60475017);安徽高校省级自然科学研究重点资助项目(KJ2007A087)

摘　　要：针对原始谱聚类初始敏感的缺点,提出了一种新的基于入侵性杂草优化(IWO)的图像聚类方法(CIWO).该算法通过计算峰值信噪比(PSNR),动态确定图像聚类簇数的最优选择范围,采用最小量差、最小簇内距离、最大簇间距离重新构造了图像聚类质量的评价函数,通过模拟杂草克隆的自然行为对图像数据集的簇中心进行快速准确定位.将算法应用于几个基准测试图像,并通过聚类有效性准则与k-Means、FCM、PSO等方法进行比较,发现CIWO具有更稳定的图像聚类性能.实验结果也表明,所提出的算法可获得更优的图像聚类质量.In order to overcome the initial sensitivity of the original spectral clustering,a novel image clustering method CIMO is presented based on the invasive seed optimization（IWO）.In this method,the optimal cluster number is dynamically determined by calculating the Peak Signal-to-Noise Ratios（PSNRs）,and a new evaluation function of clustering quality is redefined by employing the minimum quantity error,the minimum intra-instance and the maximum inter-instance.Moreover,the clustering centroids of image datasets are quickly and accurately located by simulating the natural behaviors of weed colonization.The proposed algorithm are then applied to several test benchmark images and are compared with the well-known methods such as k-Means,FCM and PSO via the clustering validation criterions.The results indicate that the proposed CIWO method is of higher clustering stability and better clustering quality.

关 键 词：模式识别 图像聚类 入侵性杂草优化 图像分析 优化 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置] TP391.41[自动化与计算机技术—控制科学与工程]