提出了一种以证据理论综合利用图像多种特征的变化检测方法。方法利用滑动窗口计算两时相图像3种特征的结构相似度, 以之构建D-S证据理论的基本概率赋值函数并进行证据合成, 通过规则判定得到图像变化区域。通过对不同试验区、不同证据组合方式以及方法间的比较实验表明, 相对单一特征检测方法有效地提高了检测的精度。此外, 由于采用统计而非原始图像特征度量特征相似性, 方法具有对辐射、几何配准精度要求较低等优点。

This paper presents an evidence theory based change detection method capable of utilizing multiple image features. With a moving window, we first get the structural similarities of both time phase image visual features and construct the basic probability assignment function (BPAF) of D-S evidence theory. We then fuse all the evidence and get the changed image areas with decision rules. Comparative work on different experimental areas, combinations of change evidence and with other meth-ods has been carried out. It shows that our method prevents effectively the detection errors from only utilizing single feature and thus improves the detection precision. Furthermore, since the image similarity is derived from image statistical features rather than original grey, texture and gradient features, this method is robust to low calibration precision.