Comparison of Change Detection Indicators in SAR Images

Change detection in multitemporal Synthetic Aperture Radar (SAR) images is a precious approach for rapid mapping applications. In this paper, we will introduce two new similarity measures for automatic SAR change detection that we compare to existing operators. These existing measures are usually based on the pixel intensity or local statistics evolution between two dates. On the one hand, the local statistics can be estimated by considering some reasonable distributions which, in practice, have been found particularly convenient in the case of SAR scenes or by using a cumulant-based series expansion. In both cases, the degree of evolution of the local statistics is measured using the Kullback-Leibler (KL) divergence. On the other hand, the measures based on pixel intensity depend on the neighboring pixel intensity inside the analysis window. In this context, we introduce two new measures for change detection in SAR images. The first one is based on pixel intensity and the second one consider the SAR local statistics. These detectors have been constructed from Rayleigh laws while supposing that the present radar texture image follows this distribution. In order to evaluate similarity measures in the comparison study, we will use two SAR images covering the same scene. Results conducted on real data illustrate the performance of these change detectors.