Change detection of remote sensing images by combining neighborhood information and structural features

In order to improve the accuracy of pixel-level change detection methods, this paper proposes a novel change detection method for remote sensing images by combining neighborhood information (including the neighborhood correlation image (NCI) and matching errors) and structural features. First, a technique of neighborhood correlation analysis is used to obtain the NCI which represents the context information, and the cross-correlation of neighborhood pixels is used to obtain matching errors by a template matching scheme. Then, structure features of images are extracted using orientated gradient information, which are robust to spectral differences between images. Subsequently, the initial change detection results is obtained by using the NCI, the matching errors, and structural features as the classification attributes of a decision tree. Finally, the Markov Random Field (MRF) is used to optimize the results, yielding the final binary map. The proposed method has been evaluated with two sets of bi-temporal remote sensing images from different sensors. Experimental results demonstrate that this method effectively improves the accuracy of change detection compared with the change vector analysis method, the single neighborhood information method and the method combining neighborhood information and texture features. © 2021, Surveying and Mapping Press. All right reserved.