A clustering approach for change detection in SAR images

This paper presents a new approach for change detection in SAR images based on clustering method. Classic change detection methods use a hard threshold to divide the difference map into two classes: change and unchanged, which has a disadvantage that some weak changed regions are often undetected. Unlike those methods, our proposed method use expectation maximization with graph cut optimization to cluster the difference map into three classes: strong changed areas, weak changed areas and unchanged areas. The experimental results on real SAR images show that our approach obtains a higher detection rate than the previous ones.