An object-based graph model for unsupervised change detection in high resolution remote sensing images
Junzheng Wu, Biao Li, Yao Qin, Weiping Ni, Han Zhang
Abstract
The difference image that represents the change levels is pivotal in unsupervised change detection tasks. An object-based graph model is proposed in this paper to generate more reliable difference images from high resolution remote sensing images. The model consists of three main steps, including segmentation, graph construction, and change measurement. First, the bi-temporal images are segmented by the fractal net evolution approach to obtain objects as the basic element for further analysis. Second, a weighted graph for each segmented object is constructed using itself and the adjacent objects as the vertexes, meanwhile, the weights are defined using objects and common boundaries. Third, a measure function is designed to evaluate the similarity between graphs, and the change level is measured based on the similarity between the graphs with the same structure in the bi-temporal images. Experimental results on three optical and two SAR datasets demonstrate the effectiveness and superiority of the proposed approach comparing with some state-of-the-art approaches.