Despeckling of SAR Image Using Generalized Guided Filter With Bayesian Nonlocal Means

摘要

Due to the coherent nature of the scattering phenomenon, synthetic aperture radar (SAR) images are inherently contaminated by multiplicative speckle noise, which makes despeckling always a fundamental problem for SAR image processing. Motivated by the idea of the guided image filter, we propose an extended despeckling scheme named generalized guided filter with Bayesian nonlocal means (GGF-BNLM). Our main contributions are as follows: 1) We successfully deduce the nonlinear weight kernel of the GGF-BNLM framework; and 2) we construct the guidance image using homogeneity analysis of local regions and the maximum-likelihood rule. Visual and quantitative experiments conducted on synthetic speckle images and real SAR images show that our method notably suppresses speckle with unperceivable detail blurring and better preserving of point-type strong scatters, which outperforms several classical and state-of-the-art methods.