JOINT CHANGE DETECTION AND IMAGE REGISTRATION METHOD FOR MULTITEMPORAL SAR IMAGES

This paper presents a novel method for jointly change detection and image registration over multitemporal Synthetic Aperture Radar (SAR) images. Image registration is performed based on histogram of unregistered images. Detecting the changes occur between the multitemporal images is performed based on the evolution of local statistics of the images. The local statistics are estimated using radon transform which removes the speckle noise occurs in the SAR images. Histogram curve fitting is used to approximate the radon probability distribution function (pdf) into Gaussian distribution which is on the assumption that radon pdf should obeys the central limit theorem. In order to measure the variation between the two pairs of projection pdf, the algorithm uses a local statistic similarity measure called Jeffrey divergence. Experiments results demonstrate that the proposed method can perform change detection rapidly and automatically over unregistered remote sensing images.