A Novel SAR Denoising Algorithm Based on NLM and Multi-Temporal SAR
Haiyan Zhang, Guoyin Cai · 2024
The most commonly used method for denoising SAR images involves applying kernel filters across the image, with filter weights typically calculated based on pixel differences within the filter kernel across multiple dimensions. While this approach effectively reduces noise and enhances image readability, it has notable drawbacks. SAR images often consist of flat regions with minimal grayscale variation and edge regions with significant grayscale differences. Conventional kernel filters tend to apply a uniform filtering approach across these different regions, which can diminish contrast near image edges and result in the loss of crucial image details. To address these issues, this paper proposes a novel method that leverages the extensive data from multi-temporal SAR images to calculate a mean image. Additionally, the Non-Local Means (NLM) filter is enhanced using a pixel block similarity criterion specifically tailored for SAR images. The improved NLM filter is then applied to the mean image to refine the denoising process of the target image. To validate the proposed method, comparative experiments were conducted using the Kuan, Lee, Frost and Bilateral filters as benchmarks, with the denoising performance evaluated through the Equivalent Number of Looks (ENL) metric. The results demonstrate that the algorithm presented in this paper achieves optimal denoising in the comparison experiments.