Published online 21 May 2008 | 453, 437 (2008) | doi:10.1038/453437a
Corrected online: 11 June 2008


No star left behind

An open-source software project could help unify every existing astronomical image into a single data set.

Left, the sky around galaxy M81; right, uses a recognized shape (top right) to line up stars in the image (red) with those in its memory (green).Left, the sky around galaxy M81; right, uses a recognized shape (top right) to line up stars in the image (red) with those in its memory (green). Click for bigger picture.J. GALLEGO

Everyone knows the experience: a pile of old photos in a dusty attic — or more likely now on a dusty disk drive — with no indication of who is in them or when they were taken. As long as the subjects are astronomical features, though, there's now an answer at hand. is an open-source software project, run out of the University of Toronto in Canada and New York University, which aims to recognize any starscape and place it in its proper coordinates within seconds — specifying not just which patch of the sky is shown, but when. By using the small relative motions of stars over time, the project's designers hope to date any picture to within a year.

The project should regularize astronomical data, making it possible to combine images from both professionals and amateurs into easily used databases. “We'd love to touch every astronomical image ever taken,” says David Hogg, a project leader and an astrophysicist at New York University. The system is currently undergoing initial alpha-testing by astronomers at various observatories, but there are plans to start accepting images from the public by September.

Although astronomers try to use a universal coordinate system to chart the sky, the reality is a messier business: telescope pointing mechanisms are idiosyncratic and not always properly used. This can make data from different observatories difficult to compare. solves the matching problem by choosing bright, four-star constellations from the pixels in a submitted image. It then looks for the same shapes in an index of 800,000 such constellations constructed from all-sky surveys catalogued by the US Naval Observatory.

In a test, the software gave the correct coordinates for all but 451 of 336,554 images from the Sloan Digital Sky Survey, a systematic map of a quarter of the sky. The team has focused on speed and reliability: all of the matches were made in less than a second, and there were no false positive matches.

Tim Axelrod, the data management project scientist for the Large Synoptic Survey Telescope, says it's the most robust calibration system he has seen, and he plans to use it when his telescope begins its surveys in 2014 (the system is powerful enough to survey the entire night sky in three days).

“The amount of money this would save the astronomical community is immense.”

As the Naval Observatory catalogues were made from multiple sky surveys done about 30 years apart, they document tiny shifts in the positions of nearby stars, called their proper motions. This means that, once has its match, it can turn back the clock, rewinding the proper motions to the best match possible for the pattern documented in a submitted image. Astronomers will thus be able to pick out features that have changed over time, such as Kuiper-belt objects in the Solar System and supernova precursors.

Jonathan Grindlay, a researcher at Harvard University who is interested in transient phenomena, is eager to use the system. He runs a programme aimed at digitally scanning more than half a million photographic plates dating back to the 1880s that, between them, cover every inch of the sky between 500 and 2,000 times. should go a long way to sorting out exactly what each plate records, and when it was taken. Once it does, a century of the skies will be open to scrutiny for changes on timescales of decades.

Eventually, Hogg would like the software to recognize any unusual feature as part of the system's routine service. But there are pressing wrinkles to iron out first. An image needs enough four-star constellations for the software to make a match, and so the system has trouble with pictures that cover less than one ten-millionth of the sky. This is an issue for images from instruments such as the Hubble Space Telescope, which looks very deeply into tiny patches of the sky. And to get the most out of the system, the software will need to match images, not just in space and time, but also in brightness — the apparent brightening and dimming of a star, for instance, would have to be referenced to something.


But as the software improves, it could become the standard for the entire community, which is trying to supply streams of data to portals such as the planned international 'virtual observatory'. Already, the US National Optical Astronomy Observatory in Tucson, Arizona, plans to be using by the end of the year to calibrate the 12 terabytes of data it collects annually from its 12 telescopes.

If it ends the laborious days of calibrating astronomical data by hand, would be saving not only lost photons, but also precious time and money, says Hogg. “The amount of money this would save the astronomical community is immense,” he says.

<b>Additional online information: Astrometry in action</b> 

See Editorial, page 428


There is a "Correction": associated with this story. The number of constellations constructed for an index in star-matching software should be 800 million not 800,000.
Commenting is now closed.