计算机应用 ›› 2011, Vol. 31 ›› Issue (11): 3004-3007.DOI: 10.3724/SP.J.1087.2011.03004

• 图形图像技术 • 上一篇    下一篇

基于空间信息的可能性模糊C均值聚类遥感图像分割

张一行1,王霞1,方世明1,李晓冬2,凌峰2   

  1. 1. 中国地质大学(武汉) 资源学院,武汉 430074
    2. 中国科学院 测量与地球物理研究所,武汉 430077
  • 收稿日期:2011-05-10 修回日期:2011-07-02 发布日期:2011-11-16 出版日期:2011-11-01
  • 通讯作者: 凌峰
  • 作者简介:

    张一行(1989-),男,湖北武汉人,主要研究方向:遥感图像处理;

    王霞(1989-),女,湖北枣阳人,主要研究方向:遥感图像处理;
    方世明(1977-),男,安徽太湖人,副教授,博士,主要研究方向:地理信息系统、地质公园规划设计;
    李晓冬(1983-),男,河北保定人,博士研究生,主要研究方向:遥感图像处理;
    凌峰(1979-),男,安徽宿松人,副研究员,博士,主要研究方向:遥感数据分析。

  • 基金资助:
    国家自然科学基金资助项目;武汉市青年晨光计划项目

Remote sensing image segmentation using possibilistic fuzzy c-means clustering algorithm based on spatial-information

ZHANG Yi-hang1,WANG Xia1,FANG Shi-ming1,LI Xiao-dong2,LING Feng2   

  1. 1. School of Resources, China University of Geosciences, Wuhan Hubei 430074, China
    2. Institute of Geodesy and Geophysics, Chinese Academy of Sciences, Wuhan Hubei 430077, China
  • Received:2011-05-10 Revised:2011-07-02 Online:2011-11-16 Published:2011-11-01
  • Contact: LING Feng

摘要: 可能性模糊C均值(PFCM)聚类算法作为模糊C均值(FCM)聚类算法的一种改进算法,能在一定程度上克服FCM算法对噪声的敏感性;但由于PFCM没有考虑像元间的空间信息,对含有较大噪声的图像分割效果依然不理想。为此,提出一种新的基于空间信息的PFCM算法(SPFCM),克服了PFCM算法对含有较大噪声的图像分割效果不佳的缺点。通过对人工图像和IKONOS遥感图像进行分析,结果表明,SPFCM算法无论是在视觉上还是在分割正确率上都优于传统的FCM算法、PFCM算法及两种加入空间信息的FCM算法;对于含有高斯噪声和盐椒噪声的图像,平均分割正确率高达99.71%,是一种去噪效果较好的图像分割算法。

关键词: 空间信息, 模糊C均值聚类, 可能性C均值聚类, 图像分割

Abstract: Fuzzy C-Means (FCM) clustering algorithm is very sensitive to image noise when it is used to image segmentation. As an improvement of FCM, Possibility FCM (PFCM) clustering algorithm can reduce the influence of image noise on image segmentation to some extent. However, since no spatial information of the image is taken into consideration, PFCM can not perform well when the image contains much noise. In order to further improve the segmentation accuracy of PFCM when much noise is present in the image, a new Spatial PFCM (SPFCM) algorithm was proposed by incorporating the spatial information of each pixel into the traditional PFCM algorithm in this paper. Both synthetic and IKONOS images with different kinds of noise were applied, and the segmentation results show that the proposed SPFCM clustering prevails over the FCM, PFCM, FCM-S1 and FCM-S2 visually and quantitatively. When dealing with different image noise, its average segmentation rate is as high as 99.71%, which shows the effectiveness of the proposed algorithm.

Key words: spatial information, Fuzzy C-Means (FCM) clustering, Possibilistic C-Means (PCM) clustering, image segmentation