Sift-ELM approach for unsupervised change detection in VHR images

This paper proposes a novel method for unsupervised change detection that is based on the Scale Invariant Feature Transform (SIFT) key points detector and the Extreme Learning Machine (ELM) classifier. The method starts by extracting SIFT key points from both images, and then matches them using the RANSAC algorithm. The matched key points based on the RANSAC algorithm are viewed as training points for the no-change class. As for the change class key points, we select them from the remaining SIFT key points extracted from the two images. The points selected are then used to train an ELM classifier. Finally, the classification map is enhanced using the Level Set segmentation algorithm. Experimental, results performed on two VHR datasets confirm the novelty of this method and its effectiveness.