EMD and KPCA-based Speckle Suppression in SAR Images

Abstract

This paper proposes a speckle suppression method for synthetic aperture radar (SAR) images based on empirical mode decomposition (EMD) and kernel principal component analysis (KPCA) following three steps: first, SAR image after logarithmic transformation is decomposed by EMD; second, noise in each intrinsic mode function (IMF) is further removed by KPCA; finally, the denoised SAR image is obtained by accumulating the IMFs processed by KPCA. In the second step, IMF, decomposed by KPCA, is reconstructed by the selection of appropriate principle components according to noise energy proportion, which is approximately calculated based on the statistical properties of speckle noise and energy distribution model of EMD-decomposed Gaussian white noise (GWN). Experimental results show that, compared with traditional EMD-based image denoising algorithms, the proposed method is superior in both speckle suppression and detail information retention.