计算机工程与应用 ›› 2016, Vol. 52 ›› Issue (10): 196-200.

• 图形图像处理 • 上一篇    下一篇

无监督特征优选高分辨率影像变化检测新方法

李小春,贾春阳,李卫华,全卫澎   

  1. 空军工程大学 信息与导航学院,西安 710077
  • 出版日期:2016-05-15 发布日期:2016-05-16

New change detection method for high-resolution remote sensing image based on unsupervised optimal selection of feature

LI Xiaochun, JIA Chunyang, LI Weihua, QUAN Weipeng   

  1. Information and Navigation College, Air Force Engineering University, Xi’an 710077, China
  • Online:2016-05-15 Published:2016-05-16

摘要: 通过分析面向对象高分辨率影像变化检测面临的问题,从影像对象多特征选择与利用入手来提高其变化检测的性能。提出了一种面向对象的非监督特征优选的变化检测方法,首先利用面向对象的分割方法对原始影像进行分割得到各影像对象并提取特征形成特征影像;然后按尽量消除数据冗余原则提取各单特征影像的变化检测结果;最后利用直方图相交的方法与基准影像中的变化强度信息作比较,并以此为依据进行特征优选。在此基础上利用马尔科夫随机场(Markov Random Fields,MRF)模型将优选特征的变化检测结果进行自动融合。新算法很好地实现了多特征的自动优选和综合利用,验证结果表明算法具有很好的变化检测准确性和鲁棒性性能。

关键词: 变化检测, 特征优选, 直方图相交, 面向对象

Abstract: Though analyzing problem of object-oriented change detection for high-resolution remote sensing image, performance of change detection will be improved by following selection and utilization of multi-feature on image’s object. An unsupervised method to select optimal feature of image’s object for change detection has been put forward. Firstly, getting image’s objects through object-oriented segmentation algorithm and forming the feature images, then acquiring results of change detection with optimal feature images on the basis of rules to eliminate redundancy from data utmostly, finally, compared with information of change intensity in the reference result through histogram cross, and get the optimal feature according to the comparing result. At last, achieve the fusion result of multi-feature using the MRF model automatically. New algorithm realizes better automatic optimal selection and integrated utilization of multi-feature, validation results show new algorithm has very good accuracy and robustness.

Key words: change detection, optimal selection of feature, histogram cross, object-oriented