Statistical Characteristic-Guided Denoising for Rapid High-Resolution Transmission Electron Microscopy Imaging

Abstract

High-Resolution Transmission Electron Microscopy (HRTEM) enables atomic-scale observation of nucleation dynamics, which boosts the studies of advanced solid materials. Nonetheless, due to the millisecond-scale rapid change of nucleation, it requires short-exposure rapid imaging, leading to severe noise that obscures atomic positions. In this work, we propose a statistical characteristic-guided denoising network, which utilizes statistical characteristics to guide the denoising process in both spatial and frequency domains. In the spatial domain, we present spatial deviation-guided weighting to select appropriate convolution operations for each spatial position based on deviation characteristic. In the frequency domain, we present frequency band-guided weighting to enhance signals and suppress noise based on band characteristics. We also develop an HRTEM-specific noise calibration method and generate a dataset with disordered structures and realistic HRTEM image noises. It can ensure the denoising performance of models on real images for nucleation observation. Experiments on synthetic and real data show our method outperforms the state-of-the-art methods in HRTEM image denoising, with effectiveness in the localization downstream task. Code will be available at https://github.com/HeasonLee/SCGN.

Figures (9)

• Figure 1: Visualization of the two kinds of statistical characteristics that we used to guide the denoising process. (a) The spatial deviation characteristic obtained by calculating the pixel standard deviation (SD) within $3\times3$ windows. It reflects the fluctuation conditions of different spatial positions in the image. (b) The frequency band characteristic obtained by the Fast Fourier Transform (FFT). Signal and noise are distributed at different frequency band positions and have different patterns.
• Figure 2: The framework of our statistical-characteristic guided network. It enhances signals and suppresses noise in the spatial and frequency domains using spatial deviation-guided weighting and frequency band-guided weighting, respectively.
• Figure 3: HRTEM image noises and their calibration. Column noise and pointwise noise can be fitted with a Gaussian distribution. The intensity of pointwise noise is proportional to the signal intensity. The intensity of column noise is almost a constant value.
• Figure 4: Denoising results on different datasets. Training datasets correspond to the testing datasets.
• Figure 5: Denoising results on real data. Gaussian filter is a traditional method. UDVD UDVD is an unsupervised learning method. The right parts of (a) and (b) are the results of supervised learning methods trained on different datasets. Red boxes mark some errors.