Paul Voyles
Discoveries and Opportunities at the Intersection of Image Science and Electron Microscopy
State of the art electron microscopy can capture digital images with a spatial resolution of 5 × 10 −11 m, easily resolving the distances between atoms in most materials, and the underlying electron scattering physics is well understood, enabling quantitative matching between experimental image data and simulations. Despite these advances, work in the field mostly still involves acquiring esthetically pleasing images that are subjected to qualitative interpretation. Methods in image science provide the opportunity to extract meaningful scientific results from noisy, distorted, and otherwise imperfect electron microscopy data. Imperfect data may be created by limitations in the experimental apparatus or by accumulating damage to the specimen as it is exposed to the electron beam.
We will present two examples of image science methods applied to electron microscopy data. First, we have used non-rigid registration to remove distortions from a series of scanning transmission electron microscopy (STEM) images, resulting in the ability to locate atoms in the images to better than 1 pm (10 −12 m) precision [1]. We have used high precision STEM to image the distortions of the surface atoms on Pt nanocatalysts that are responsible for their catalytic function and to image vacancies, one of the last point defects in crystals to elude detection by atomic-resolution imaging.
Second, we have adapted modern non-local denoising methods to STEM data. STEM data at low dose is typically limited by Poisson noise, not additive Gaussian noise, so off-the-shelf methods are not effective at very low signal levels. We have adapted the BM3D method to account for Poisson noise and to exploit the periodic nature of most STEM images for computational efficiency [2], and combined non-rigid registration with non-local principle component analysis denoising [3]. These methods enalbed recovery of the site occupied by an impurity atom in a crystal from data with fewer than ten counts per pixel in the original data.
We will conclude with an outlook on future opportunities at the intersection of image science and electron microscopy [4]. These opportunities include developing machine learning methods to discover useful signals in the large data sets that will be created by new detector technologies and automated optimization methods to derive atomic structure information from electron microscopy data with dramatically reduced human intervention.
Realizing these opportunities will require continued close collaboration between applied mathematicians and domain scientists in microscopy, similar to the examples above.
References
[1] A. B. Yankovich, B. Berkels, W. Dahmen, P. Binev, S. I. Sanchez, S. A. Bradley, A. Li, I. Szlufarska, and P. M. Voyles, Picometre-precision analysis of scanning transmission electron microscopy images of platinum nanocatalysts. Nature Communications 5:4155, 2014.
[2] N. Mevenkamp, P. Binev, W. Dahmen, P. M. Voyles, A. B. Yankovich, and B. Berkels, Poisson noise removal from high-resolution STEM images based on periodic block matching. Advanced Structural and Chemical Imaging 1:3, 2015.
[3] A. B. Yankovich, C. Zhang, A. Oh, T. J. A. Slater, F. Azough, R. Freer, S. J. Haigh, R. Willett, and P. M. Voyles, Non-rigid registration and non-local principle component analysis to improve electron microscopy spectrum images. Nanotechnology 27:364001, 2016.
[4] P. M. Voyles, Informatics and data science in materials microscopy, Current Opinion in Solid State and Materials Science, DOI: 10.1016/j.cossms.2014.02.003