Automatic Change Detections from SAR Images Using Fractal Dimension

It is very difficult to detect changes from SAR images because of two major difficulties associated with SAR, which are the removal of speckle noise and the registration of information between images. Speckle is a chaotic phenomenon because that the scattering signals within a resolution cell are summed up coherently. Therefore, SAR signal can be modeled by a spatial chaotic system and characterized by its fractal dimension. Then, simplified procedures for SAR image change detection are proposed because that the process of image despeckling is unnecessary. The proposed approach is applied to multitemporal polarimetric SAR images for change detections. The experimental results of using a simple image difference (DI) technique, the principal component analysis (PCA), and the proposed (Fractal) approach are compared. The effects of misregistration for different approaches are also presented. Simulation results reveal that misregistration affects less and less as SNR is increased. When SNR is low, by using DI or PCA methods, the overall performance of change detection is degraded by spurious differences due to misregistration. On the contrary, Fractal method can tolerate misregistration effect at low SNR. In addition, when the difference between changed classes is small, it is fail to detect changes by using of either DI or PCA method. In contrast, the Fractal method can still effectively detect land cover changes.