Non-local Means for Scanning Transmission Electron Microscopy Images and Poisson Noise based on Adaptive Periodic Similarity Search and Patch Regularization

Abstract

High-Angle Annular Darkfield Scanning Transmission Electron Microscopy (HAADF-STEM) allows to take images at atomic scale with a contrast proportional to the atomic number. STEM acquires an image line-by-line, pixel-by-pixel leading to characteristic distortions. Furthermore, STEM images of beam sensitive materials have to be taken with short exposure times, leading to low contrast images with Poisson noise. In this paper, we propose an extension of Non-local Means (NLM) tailored to STEM images of crystalline structures. To find similar patches, we introduce an adaptive non-local search strategy that exploits the periodic structure of the crystal images. Furthermore, we extend the patch similarity measure to take into account the horizontal distortions typical for STEM images. Moreover, we discuss the Anscombe transform and the Poisson likelihood ratio to deal with Poisson noise. Finally, the resulting methods are compared to BM3D with Anscombe tranform and PURE-LET on simulated and real data.