SAR image change detection using regularized dictionary learning and fuzzy clustering

In this paper, we propose and present a novel unsupervised change detection(CD) algorithm for synthetic aperture radar(SAR) images based on regularized dictionary learning and fuzzy clustering. The regularized sparse reconstruction technique is introduced to generate a de-noised, low time consuming reconstructed image by using K-SVD dictionary learning. In order to obtain proper difference image, minus and ratio maps are discussed with the comparison of the other state-of-the-art approaches. Finally, to transfer the difference map into change map, we employ the optimized FCM called FLICM algorithm to undertake the task which aims to segment the difference map into two classes: changed and unchanged. Experimental results clearly show that the proposed approach consistently yields superior performance (accuracy, efficiency and robustness) compared to several well-known change detection techniques on both noise-free and noisy satellite images, further optimization methods are discusses in the end.