Patch Ordering-Based SAR Image Despeckling Via Transform-Domain Filtering

Abstract

In this paper, we propose a synthetic aperture radar (SAR) image despeckling method based on patch ordering and transform-domain filtering. Logarithmic transformation with bias correction is applied to the original SAR image to transform the multiplicative noise model into the additive model. Then, we adopt a two-stage filtering strategy. The first stage is coarse filtering which can suppress speckle effectively. In this stage, we extract the sliding patches from the logarithmic SAR image, and order them in a smooth way by a simplified patch ordering algorithm specially for SAR images. The ordered patches are filtered by learned simultaneous sparse coding (SSC), a technology recently advanced in image processing. Then, the coarse filtering result is reconstructed from the filtered patches via inverse permutation and subimage averaging. The second stage is refined filtering which can eliminate small artifacts generated by the coarse filtering. In this stage, the sliding patches are extracted from the coarse filtering result and ordered in the same way. Then, we apply 2-D wavelet hard-thresholding to the ordered patches and reconstruct the refined filtering result. The final result is obtained by taking exponential transformation to the refined filtering result. An algorithm based on the proposed strategy is presented in detail and the parameters are selected for fast and effective realization. Experimental results with both simulated images and real SAR images demonstrate that the proposed method achieves state-of-the-art despeckling performance in terms of peak signal-to-noise ratio (PSNR), structural similarity (SSIM) index, equivalent number of looks (ENLs), and ratio image.