测绘学报

• 学术论文 • 上一篇    下一篇

基于感兴趣区域在区域层面上决策的SAR图像变化检测的方法研究

万红林1,2,焦李成2,王桂婷2,辛芳芳2   

  1. 1.
    2. 西安电子科技大学
  • 收稿日期:2010-08-27 修回日期:2011-05-30 出版日期:2012-04-25 发布日期:2012-04-25
  • 通讯作者: 万红林

A Technique of Decision at the Region Level Based on Region of Interest for Change Detection in SAR Images

  1. 1.
    2. Key Laboratory of Intelligent Perception and Image Understanding of Ministry of Education of China, Xidian University, Xi'an 710071, P.R. China
  • Received:2010-08-27 Revised:2011-05-30 Online:2012-04-25 Published:2012-04-25

摘要: 在现有的变化检测方法中,针对差分图像的方法得到了广泛的应用,但现有该框架下的技术都在在像素层面上决策产生的变化检测结果,这易使结果中存在诸如有噪声般散落的杂点、连通区域内有孔洞及边缘不平滑等缺陷。为此,本文给出一种在区域层面上决策生成变化检测结果的技术,其核心为抽取并处理感兴趣区域,而关键在于获取合适的抽取感兴趣区域的标签和如何在区域层面上生成变化检测结果。为使抽取的感兴趣区域包含几乎所有的变化类信息,我们用平稳小波变换和模糊C-均值(Fuzzy C-Means, FCM)算法分两步获取抽取感兴趣区域的标签;为在区域层面上生成变化检测结果,我们依据标签搜索感兴趣区域内所有的连通区域,并把每个连通区域看作为一个数据点,再由阈值技术处理这些数据点生成最终的变化检测结果。对真实SAR图像数据集的变化检测结果表明,其主观效果和客观性能都优于其他相关技术的。

Abstract: In the context of change detection in SAR images, most of techniques are based on the analysis of the difference image, while almost all make the decision at the pixel level. This decision way will cause the change detection map noisy, with holes in connected component and jagged boundaries. For avoiding this, a technique making the decision at the region level is proposed, whose core is extracting and handling region of interest (ROI) and the key is obtaining the proper the label used to guide for extracting ROI and that how to make the result generated at the region level. To make the extracted ROI contain nearly all pixels in the changed area, stationary wavelet transform (SWT) and fuzzy c-means (FCM) algorithm have been used to get it; to make the change detection map generated at the region level, all connected regions in the extracted ROI are searched and each one is looked at as a unit to be handled. Then the threshold-based method is used to group those units into two classes. Results tested on the real SAR datasets have shown that our method outperforms other related techniques, whatever from quantitative or subjective aspects.