基于改进谱聚类与粒子群优化的图像分割算法  

Image Segmentation Algorithm Based on Improved Spectral Clustering and Particle Swarm Optimization

作　　者：杨锋[1] 柴毅[2] 

机构地区：[1]重庆大学自动化学院,重庆400044 [2]输配电装备及系统安全与新技术国家重点实验室,重庆400044

出　　处：《微电子学与计算机》2013年第7期51-54,59,共5页Microelectronics & Computer

基　　金：国家自然科学基金项目(60974090);重庆市科技攻关项目(cstc2010ac3055)

摘　　要：本文提出了一种基于改进谱聚类与粒子群优化的图像分割算法.该算法利用双树复小波变换系数,求得能量均值构造相似性矩阵,充分利用了待聚类数据所包含的空间邻近信息和特征相似性信息.在谱映射的过程中,采用了Nystrm逼近策略,降低了谱聚类算法的复杂度和内存消耗,然后在进行K均值聚类时使用粒子群优化算法.最后,通过对医学图像和遥感图像分割验证了新算法的有效性.A image segmentation algorithm based on improved spectral clustering and particle swarm optimization is proposed.Similarity matrix is constructed by the mean of dual-tree complex wavelet transform coefficients in this dissertation so as to make full use of the spatial adjacency information and feature similarity information included in the data.To efficiently apply the algorithm to image segmentation,Nystrm approximation strategy is used in the course of spectral mapping to reduce the computation complexity and memory consumption.And then we tentatively adopt particle swarm optimization algorithm to optimize the K-means clustering in the spectral clustering algorithm.Experimental results on medical images and remote sensing images verify the validity of the proposed algorithm.

关 键 词：图像分割 粒子群 谱聚类 Nystrom逼近 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]