基于特征空间聚类结构信息的遥感影像边缘增强方法  

A method to enhance edges of remotely sensed imagery based on cluster structure information in feature space

作　　者：魏志华[1] 毛政元[1,2] 

机构地区：[1]福州大学空间数据挖掘与信息共享教育部重点实验室,福建福州350002 [2]福州大学福建省空间信息工程研究中心,福建福州350002

出　　处：《山东大学学报（工学版）》2011年第5期76-81,共6页Journal of Shandong University（Engineering Science）

基　　金：国家自然科学基金资助项目(40871206);国家科技支撑计划资助项目(2007BAH16B01)

摘　　要：为解决边缘增强效果依赖于模板选择的问题,提出了一种新的彩色图像边缘增强方法,该方法以高空间分辨率遥感影像在红、绿、蓝三波段构成的三维特征空间中的聚类中心数目、位置以及分布模式信息为基础,在不同聚类层次上通过缩短同一聚类中各点到其集聚中心的距离,扩大类间距,减小类内差异,实现增强不同尺度目标区域边缘,抑制背景区域和平滑噪声的目标。实验结果证明,该方法能明显增强多波段遥感影像的边缘信息。A novel method for color image edge enhancement was proposed in order to solve the problem that the effect of edge enhancement depends on the chosen templates,which was based on the extracted information of number,location and distribution pattern of cluster centers in a three-dimensional feature space constructed by red,green and blue bands of high spatial resolution remotely sensed imagery.Enhancing edges of multi-scale target areas,suppressing background areas and smoothing noise were achieved by shortening distances of each point to the cluster center it belongs to at different cluster levels,in order to expand distances among different classes and reduce their internal discrepancies.Experimental results proved that the algorithm can effectively enhance edges of multispectral remotely sensed imagery.

关 键 词：高空间分辨率遥感影像 边缘增强 聚类分析 几何概率 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP751[自动化与计算机技术—计算机科学与技术]