A Novel Superpixel Cosegmentation for Change Detection in Remote Sensing Imagery

Abstract

In this paper, a novel superpixel cosegmentation framework for Change Detection (CD) is proposed. First, simple linear iterative clustering is implemented to bi-temporal images to get superpixel maps. Based on multivariate probability density functions of the corresponding superpixels in two maps, a similarity map is then measured by multivariate Kullback-Leibler distance to represent the change feature. Next, combined with the respective image features of the bi-temporal images, two different detection results are obtained by energy minimization using a superpixel graph cut algorithm. Finally, by comparing the relationship between the changed objects in two different CD maps, the final change result is obtained. And the experiment results of high spatial resolution dataset demonstrate the effectiveness and superiority of the proposed algorithm.