Robust Registration of Cloudy Satellite Images Using Two-Step Segmentation

In this letter, we propose an effective registration method for cloudy satellite images based on global and local thresholds. First, cloud candidates are determined by using optimal threshold and κ-means clustering. Then, using the local threshold, the cloud candidates are further classified into three categories: thick clouds, thin clouds, and ground. Finally, accurate registration is performed by eliminating features relating to cloudy areas. The experiments show that the proposed method provides segmentation accuracy of 93.29%. In addition, registration accuracy is improved by 24.83%, as compared with conventional methods.