Remote Sensing Image Denoising Using Patch Grouping-Based Nonlocal Means Algorithm

• Published in: IEEE geoscience and remote sensing letters Vol. 14; no. 12; pp. 2275 - 2279
• Main Authors: Xu, Su, Zhou, Yingyue, Xiang, Hongbing, Li, Shuiying
• Format: Journal Article
• Language: English
• Published: Piscataway IEEE 01.12.2017; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Algorithms; Blockwise; Blurring; Estimation; Filtration; Image denoising; Image edge detection; Mathematical models; Noise reduction; nonlocal means (NLM); patch grouping; Patches (structures); Preservation; Remote sensing; remote sensing image; Time complexity
• ISSN: 1545-598X; 1558-0571
• DOI: 10.1109/LGRS.2017.2761812

Remote sensing images contain repetitive image patches, which makes nonlocal means (NLM) algorithm particularly suitable to denoise them. Blockwise NLM (BNLM) improves NLM's shortage of high time complexity, but still has the problems of edge blurring and details losing. A patch grouping-based NLM (NLMPG) algorithm is proposed in this letter, and it follows BNLM in estimating the value of a patch by its similarity with other patches in the image, but it improves in two aspects: first, instead of using all image patches in the search window to denoise the center patch, only numbers of the most similar patches are selected, which helps to get rid of less relevant information, and second, NLM and BNLM use the same filtering constant for the whole image, but NLMPG customizes the value of filtering constant for each center patch by ratio of image patch variances, resulting in better performance. Experimental results verify that the proposed NLMPG algorithm is good at structure maintenance and edge preservation, achieving the state-of-the-art denoising performance in terms of both quantitative criteria and subjective visual quality.