Multi-image Overlap Stitching and Automatic Image Construction for Coherent X-ray Imaging
This paper introduces MOSAICX, an automated workflow that efficiently stitches hundreds of partially overlapping direct-space coherent X-ray images into a single large-area composite, enabling the comprehensive visualization and analysis of magnetic-domain structures beyond the field of view of individual measurements.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
To see the invisible architecture of a material, scientists often turn to X-rays, a form of light so energetic it passes through solid matter. But when these X-rays are coherent—meaning their waves are perfectly synchronized like soldiers marching in step—they can reveal details far smaller than a single atom. This technique, known as coherent X-ray imaging, allows researchers to map the magnetic landscapes inside materials without touching them. In a specific class of materials called antiferromagnets, the internal magnetic order is hidden because the tiny magnetic arrows point in opposite directions, canceling each other out on a large scale. Yet, at the boundaries where these opposing directions meet, known as domain walls, the cancellation is imperfect, creating distinct signatures that can be imaged. Understanding these boundaries is crucial because they dictate how the material conducts electricity and responds to magnetic fields, properties that could power the next generation of ultra-fast, energy-efficient electronics. However, a single snapshot from an X-ray camera captures only a tiny fraction of the material, leaving the vast majority of the magnetic landscape unseen.
Researchers at the University of Wisconsin–Milwaukee have developed a new method to solve this problem of scale, allowing them to stitch together hundreds of tiny, overlapping snapshots into one massive, coherent picture. The team focused on a crystal of manganese bismuth telluride, a material that acts as both a topological insulator and an antiferromagnet. When they scanned the surface of this crystal, the X-ray camera captured 434 individual images, each showing a small patch of the magnetic domain walls as dark, wavy lines against a bright background. The challenge was that these images were not perfectly aligned; slight shifts in the sample's surface and the geometry of the X-ray beam meant that the edges of one image did not naturally line up with the next. Manually aligning hundreds of images would be tedious and prone to human error, potentially introducing distortions that would ruin the scientific data.
To overcome this, the team created an automated workflow they call MOSAICX. The process begins by treating each of the 434 images individually. The computer first finds the exact center of the circular light pattern in every photo, correcting for any slight shifts in position. It then creates a "mask," which is essentially a template of the background noise and static imperfections common to that specific set of scans. By subtracting this background template from each image, the software isolates the magnetic domain walls, making them stand out clearly. The images are then trimmed to remove the fuzzy edges and converted into a simple black-and-white map where the domain walls are solid black lines and everything else is white. This simplification is key, as it strips away distracting details and leaves only the structural features needed for alignment.
Once the images are cleaned up, the software begins the stitching process in two steps. First, it takes the images from a single row of the scan and aligns them side-by-side. It does this by sliding the images over one another and calculating how well the black domain-wall lines overlap. When the lines match up perfectly, the software locks that position and merges the images into a single long strip. This is repeated for all fourteen rows of the scan, creating fourteen long columns. The second step involves stitching these fourteen columns together vertically. The same overlap-matching logic is applied, sliding the columns until the magnetic patterns align seamlessly across the entire surface. The researchers found that sometimes, if two images had very few matching features, the computer might make a mistake and skip a piece or align it incorrectly. To handle this, they built in a check where they could manually remove any faulty columns and re-run the stitching, ensuring the final map was accurate.
The final result is a single, massive image that reveals the complete magnetic domain landscape of the scanned area, a view that is far larger than what any single camera could capture. The team verified the quality of their work by looking at the edges where the images joined. They found that while the initial stitched map accurately showed the large-scale network of magnetic walls, the fine details of the magnetic walls, including the subtle ripples and intensity changes that occur where the X-rays interfere with each other, were lost during the simplification process. To recover these details, the researchers used the precise coordinates determined by the automated stitching to reconstruct a high-fidelity composite image directly from the original raw data. This final reconstruction showed a continuous network of magnetic walls stretching across the material, free from the gaps, double edges, or detector defects that usually plague such large-scale reconstructions. This method proves that it is possible to automate the assembly of complex scientific data, turning a fragmented collection of hundreds of small pictures into a unified, high-resolution map of a material's hidden magnetic world.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.