Benchmarking featurisation methods for segmenting HRTEM images

High-Resolution Transmission Electron Microscopy (HRTEM) is critical for controlling material synthesis for modern high-efficiency precision applications, as it enables atomic-scale characterisation of material microstructure. However, identifying different structural regimes of order and disorder still relies on manually-tuned image processing or bespoke and computationally demanding supervised-learning approaches. We benchmark the performance of different textural, Fourier, topological, and distributional features for crystalline-region segmentation of experimental HRTEM images.