A New Model-Independent Method for Change Detection in Multitemporal SAR Images Based on Radon Transform and Jeffrey Divergence

This letter presents a new approach for change detection in multitemporal synthetic aperture radar images. Considering about the existence of speckle noise, the local statistics in a sliding window are compared instead of pixel-by-pixel comparison. Edgeworth series expansion is applied to estimate the probability density function (pdf), which is on the assumption that the pdf is not too far from normal distribution. To transcend such a limitation, in each analysis window, the image is projected onto two vectors in two independent dimensions; thus, the pdf of each projection is closer to a Gaussian density. In order to measure the distance between the two pairs of projections, the proposed algorithm uses a modified Kullback–Leibler (KL) divergence, called Jeffrey divergence, which turns out to be more numerically stable than KL divergence. Experiments on the real data show that the proposed detector outperforms all the others when a high detection rate is demanded.