Kuan-modified speckle reducing anisotropic diffusion: A hybrid Bayesian/non-Bayesian method for despeckling SAR...

Speckle is a form of granular noise that influences SAR image quality and hinders the extraction of radiometric/textural features. Traditional methods for reducing such noise involve a trade-off between noise reduction and edge preservation. The advantage of Bayesian filters such as Kuan filter is the ability to smooth out homogeneous regions while the disadvantage is the blurring of the edges, whereas non-Bayesian, PDE-based filters such as anisotropic diffusion (AD) perform well on the edges despite being computationally inefficient and unsuitable for multiplicative noise. In this article, the step-by-step method of developing a hybrid algorithm for combining Kuan MMSE and anisotropic diffusion filters via window-based classification via the conduction function of Speckle Reducing Anisotropic Diffusion (SRAD) method is described. Procedure and all the equations as well as an application example of comparing four filters (one of which is developed) applied to two test SAR images for the speckle variance range of 0-0.1 are given. • Despeckling method involving hybrid approach of Kuan MMSE and anisotropic diffusion filters based on the conduction function of SRAD. Provides explicit analytical expressions for all filter parameters (effective number of looks, image/speckle CoVs, weight coefficient, and edge threshold). • Compares PSNR/SSIM performance with Lee, Frost, Kuan (adaptive), and Perona-Malik filters for 0 to 0.1 speckle variance range in two different SAR test images.