Unsupervised change detection using fuzzy c-means and MRF from remotely sensed images

In this paper, a novel change detection approach is proposed using fuzzy c-means (FCM) and Markov random field (MRF). First, the initial change map and cluster (changed and unchanged) membership probability are generated through applying FCM to the difference image created by change vector analysis (CVA) method. Then, to reduce the over-smooth results in the traditional MRF, the spatial attraction model is integrated into the MRF to refine the initial change map. The adaptive weight is computed based on the cluster membership and distances between the centre pixel and its neighbourhood pixels instead of the equivalent value of the traditional MRF using the spatial attraction model. Finally, the refined change map is produced through the improved MRF model. Two experiments were carried and compared with some state-of-the-art unsupervised change detection methods to evaluate the effectiveness of the proposed approach. Experimental results indicate that FCMMRF obtains the highest accuracy among methods used in this paper, which confirms its effectiveness to change detection.