Improved nonlocal means algorithm for synthetic aperture radar image despeckling based on optical guidance

To address the challenge of simultaneously enhancing detail preservation and noise suppression in synthetic aperture radar (SAR) image denoising, we propose an optically guided modified nonlocal means (NLM) method. Utilizing the BigEarthNet v2.0 dataset as a benchmark, the method employs Sentinel-2 optical imagery to guide the denoising of Sentinel-1 SAR images. The proposed approach incorporates an adaptive search window selection module designed based on edge density information derived from the optical image. This module dynamically adjusts the search window size according to the edge density value of each pixel to better preserve structural details. Furthermore, a similarity measurement function is defined to compute similarity between distinct local patch regions, leveraging the gamma distribution characteristics inherent in speckle noise. For the calculation of the smoothing parameter h, distinct computational strategies are applied by integrating local noise intensity and edge density values. This dual strategy enables stronger noise suppression in homogeneous areas and superior detail retention in edge regions. Experimental results demonstrate that the proposed algorithm achieves enhanced radiometric consistency and edge preservation capability compared with other benchmark methods while maintaining robust speckle noise suppression. It exhibits particularly strong performance in scenes characterized by high land cover complexity.