Local low rank denoising for enhanced atomic resolution imaging.

Area of Science:

• Materials Science
• Physics
• Data Science

Background:

• Atomic resolution imaging and spectroscopy are limited by low signal-to-noise ratios, hindering detailed analysis.
• Interpreting individual pixels or spectra in such data is often challenging due to noise.

Purpose of the Study:

• To introduce and evaluate local low rank (LLR) denoising for noise reduction in scanning transmission electron microscopy (STEM) and electron energy-loss (EEL) spectrum images.
• To demonstrate the efficacy of LLR denoising in preserving fine structural details and revealing subtle spectral features.

Main Methods:

• Utilized tensor decomposition, specifically multilinear singular value decomposition (MLSVD), for LLR denoising.
• Applied LLR denoising to STEM images of graphene and EEL spectrum images of CoFe2O4.
• Assumed the signal of interest is of low rank in appropriately sized data segments.

Main Results:

• LLR denoising successfully suppressed statistical noise in graphene STEM images.
• The method preserved fine image features, including scan row-wise distortions related to graphene rippling.
• LLR denoising revealed fine spectral structures in EEL data, differentiating lattice sites in CoFe2O4.

Conclusions:

• LLR denoising is an efficient and versatile tool for noise removal in atomic resolution microscopy.
• The technique significantly improves the interpretability of STEM images and EEL spectra.
• LLR denoising offers a general approach for enhancing scientific data quality, independent of signal features or data dimensions.