The paper introduces a new denoising method called non-local means (NL-means). The method works by replacing the color of a pixel with an average of the colors of similar pixels. Unlike traditional methods that only consider nearby pixels, NL-means searches the entire image to find the most similar pixels, which may not be close. This approach is more effective in preserving image details, especially in areas with periodic patterns or elongated edges. The algorithm evaluates resemblance by comparing color patches around each pixel, not just individual pixel colors. The resemblance is measured using a Euclidean distance between patches, and a decreasing function is applied to weight the similarity. The weight function ensures that patches with smaller distances (indicating higher similarity) contribute more to the average. The paper presents two implementations: a pixelwise and a patchwise version. The pixelwise version averages colors within a local neighborhood, while the patchwise version considers larger patches and averages results across multiple patches. The patchwise implementation generally provides better PSNR (peak signal-to-noise ratio) due to more effective noise reduction, though it does not necessarily improve detail preservation. The method's performance depends on parameters such as the patch size, search window size, and filtering parameter h, which is set as h = kσ, where σ is the noise standard deviation and k is a constant. Larger patches and search windows are needed for higher σ values to ensure robustness and better noise removal. The algorithm is implemented in ANSI C and is available online. Some files may be subject to patent restrictions, and users are advised to check the licensing and patent information before using them. The rest of the files are distributed under the GPL license. The method is effective in removing noise while preserving image details, as demonstrated in the example provided.The paper introduces a new denoising method called non-local means (NL-means). The method works by replacing the color of a pixel with an average of the colors of similar pixels. Unlike traditional methods that only consider nearby pixels, NL-means searches the entire image to find the most similar pixels, which may not be close. This approach is more effective in preserving image details, especially in areas with periodic patterns or elongated edges. The algorithm evaluates resemblance by comparing color patches around each pixel, not just individual pixel colors. The resemblance is measured using a Euclidean distance between patches, and a decreasing function is applied to weight the similarity. The weight function ensures that patches with smaller distances (indicating higher similarity) contribute more to the average. The paper presents two implementations: a pixelwise and a patchwise version. The pixelwise version averages colors within a local neighborhood, while the patchwise version considers larger patches and averages results across multiple patches. The patchwise implementation generally provides better PSNR (peak signal-to-noise ratio) due to more effective noise reduction, though it does not necessarily improve detail preservation. The method's performance depends on parameters such as the patch size, search window size, and filtering parameter h, which is set as h = kσ, where σ is the noise standard deviation and k is a constant. Larger patches and search windows are needed for higher σ values to ensure robustness and better noise removal. The algorithm is implemented in ANSI C and is available online. Some files may be subject to patent restrictions, and users are advised to check the licensing and patent information before using them. The rest of the files are distributed under the GPL license. The method is effective in removing noise while preserving image details, as demonstrated in the example provided.