Non-Rigid Registration for STEM

This tool provides non-rigid registration and averaging of a series of scanning transmission electron microscopy images.

Launch Tool

You must login before you can run this tool.

Version 1.0 - published on 02 Dec 2015

doi:10.4231/D30R9M519 cite this

This tool is closed source.

View All Supporting Documents



Published on


This tool takes as inputs a series of fast scanning transmission electron microscopy (STEM) images of the same object. It applies a non-rigid registration algorithm to correct for sample drift between frames and scan instabilities within each frame. Then it returns the average image over the entire series of frames, an image of the number of frames used for each pixel in the average, and the entire data series after registration. Pixels that fall outside the frame after registration due to drift are set to zero. The averaged frame has improved signal to noise ratio and reduced scan distortions with minimal loss of spatial resolution, potentially achieving sub-picometer precision in atom position locations from atomic-column resolved STEM images. STEM images for registration should be acquired with short enough dwell time to provide adequate sampling of microscope instabilities (10-12 microseconds per pixel) and should be acquired without line synchronization of the scan.

Usage example
1.    Click on “Use default datafile.”
2.    After image conversion, several preview images should appear in the Image Preview box.
3.    Click on “Submit to cluster.”
4.    After several minutes, an email should appear in the email account associated with your Nanohub account.
5.    Click on “Download results” to download the <input name>_<DATE>_<TIME>_results.tar file.  The tar file contains preview images in 8-bit PNG format and true quantitative results in floating-point TIF files.

Starting pixel size (Angstroms/pixel side)
This field controls the startLevel parameter.

In order to avoid unwanted local minima, the NRR algorithm does a coarse to fine registration, downsampling the original image to produce a coarse image, performing registration on that coarser image, and proceeding to the next finer level using the shift produced in the coarse level.

The recommended resolution is a pixel size which will produce startLevel = stopLevel – 1.
In most cases, the resolution should not be coarser than 1.7 Angstroms (170 pm) per pixel side, and startLevel should be no less than stopLevel – 4.

TEM images have an image side length of 2^n. Assuming an input image of 256*256 pixels, or 2^8 * 2^8, stopLevel is defaulted as 8, and the starting pixel size is defaulted as 0.2 Angstroms per pixel side, to produce a startLevel of 7.




Thank you to Chenyu Zhang, Jie Feng, and Dr. Clark at nanoHUB for assistance during tool creation.


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, doi:10.1038/ncomms5155 (2014).

B. Berkels, P. Binev, D. A. Blom, W. Dahmen, R. C. Sharpley and T. Vogt, Optimized imaging using non-rigid registration. Ultramicroscopy 138, 46–56 (2014).

Cite this work

Researchers should cite this work as follows:

  • Tam Mayeshiba, Weixi Ma, Henry Wu, Benjamin Berkels, Andrew B. Yankovich, Wolfgang Dahmen, Peter Binev, Dane Morgan, and Paul M. Voyles, "Non-Rigid Registration for STEM," (2015).

  • (2015), "Non-Rigid Registration for STEM," (DOI: 10.4231/D30R9M519).

    BibTex | EndNote