Stripe Noise Removal in Scanning Probe Microscopy

For more details on the code, please refer to the ReadMe file. Here you will find the description of the paper: To find the most robust and efficient method for removing stripe noise from c-AFM images, we developed a noise model and performed intensive comparison on noisy c-AFM images and simulated noisy images. Most image data that support the findings of this study are openly available in [DataverseNL] at [https://doi.org/10.34894/OYIGPC], reference number [1110]. The comparison of the 16 selected methods includes 1) Low-Rank Recovery, Group Sparse Recovery, and Unidirectional Total Variation Minimization; 2) all denoising methods using line-by-line scanning in Gwyddion; 3) a deep learning method developed and trained for stripe noise removal only; 4) two state-of-the-art denoising methods developed for AFM images. This code is divided into two: Visual Comparison, and Quantitative Image Quality Comparison. We analyze the results of natural noise removal using all 16 methods. In the subsection Quantitative Image Quality Comparison, we propose our noise model to simulate noisy images and perform two experiments. We propose the noise model and verify if the visual and SSIM results on simulated noisy results and visual results on natural noisy images are consistent to check the validity of the noise model. The first Quantitative Image Quality Comparison experiment uses a set of images that have very different noise strengths, and the other uses an image dataset with random noise strengths. In the second experiment, we create a dataset of 800 ground truths by flipping and cropping a clean image and adding random simulated stripe noise to obtain a corresponding dataset of 800 simulated noisy images. We obtain the SSIM curve and PSNR (Peak Signal to Noise Ratio) curve from the first experiment and the boxplots of the 800 PSNR and SSIM results from the second experiment, both leading to the same conclusion regarding the best denoising method