Materials property mapping from atomic scale imaging via machine learning based sub-pixel processing

This paper presents a machine learning-based sub-pixel processing method for mapping materials properties from atomic-scale scanning transmission electron microscopy (STEM) images. The method involves primary segmentation of atomic signals from background noise, followed by a denoising process using block matching and 3D filtering with Anscombe transformation and morphological filtering. K-means clustering is used for robust thresholding to extract atomic column centroids, achieving sub-pixel accuracy. The method is benchmarked against other existing STEM analysis programs, demonstrating improved accuracy and computational efficiency.

Junghun Han, Kyoung-June Go, Jinhyuk Jang, Sejung Yang, Si-Young Choi