基于滤波器的聚类算法  

A Filter-Based Clustering Algorithm

作　　者：张强[1] 李成[1] 吴腾飞[1] 

机构地区：[1]天津大学精密测试技术及仪器国家重点实验室,天津300072

出　　处：《天津大学学报》2010年第10期884-889,共6页Journal of Tianjin University(Science and Technology)

基　　金：中国博士后科学基金资助项目(20090450767);天津市高等学校科技发展基金资助项目((20080810)

摘　　要：针对聚类分析精度和效率低的问题,设计了一种聚类算法FBCLUS.应用卷积定理和傅里叶变换,提出了频率滤波法来消除噪声的干扰;提出了单阈值、多阈值幅度滤波法消除噪声和提取不同密度的感兴趣区间;提出一个数学形态学算子提取聚类簇.实验表明:FBCLUS算法能够发现任意形状的聚类;速度快,计算复杂度为O(N);能够发现不同密度的聚类簇;抗噪声性能强;对网格大小有一定的适应性.FBCLUS算法有很高的聚类精度和效率.To overcome the problems of low accuracy and low efficiency,a clustering algorithm FBCLUS was proposed. Firstly, convolution theorem and Fourier transform were used to design frequency filters to reduce noise; single threshold and multi-threshold amplitude filters were introduced to reduce noise and distinguish different density regions of interest. Thirdly, a mathematical morphology clustering operator was designed to discover clusters. The experiments show that FBCLUS is able to detect arbitrarily shaped clusters, and it is very efficient with a complexity of O（N）; it can distinguish clusters of different density; it is insensitive to large amounts of noise; and it is not sensitive to the grid size. FBCLUS has high accuracy and efficiency.

关 键 词：频率滤波 幅度滤波 数学形态学 聚类 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]