Conductive atomic force microscopy (c-AFM) can provide simultaneous maps of the topography and electrical current flow through materials with high spatial resolution and it is playing an increasingly important role in the characterization of novel materials that are being investigated for novel memory devices. However, noise in the form of stripe features often appear in c-AFM images, challenging the quantitative analysis of conduction or topographical information. To remove stripe noise without losing interesting information, as many as sixteen destriping methods are investigated in this paper, including three additional models that we propose based on the stripes characteristics, and thirteen state-of-the-art destriping methods. We have also designed a gradient stripe noise model and obtained a ground truth dataset consisting of 800 images, generated by rotating and cropping a clean image, and created a noisy image dataset by adding random intensities of simulated noise to the ground truth dataset. In addition to comparing the results of the stripe noise removal visually, we performed a quantitative image quality comparison using simulated datasets and 100 images with very different strengths of simulated noise. All results show that the Low-Rank Recovery method has the best performance and robustness for removing gradient stripe noise without losing useful information. Furthermore, a detailed performance comparison of Polynomial fitting and Low-Rank Recovery at different levels of real noise is presented.
© 2024. The Author(s).