Making the sky searchable
Computer scientists at the University of Toronto (U of T) have teamed up with astronomers at New York University on an ambitious project. You can send them a picture of the sky above your head and their special software will identify the stars that are in the image. In other words, their computer program will make night sky searchable. The team is organizing and mixing images coming from astronomical databases with images coming from 'all kinds of cameras, amateur telescopes, large ground-based telescopes, and space telescopes such as the Hubble Space Telescope.' This specialized search engine is still in beta-version, but is available to both professional and amateur astronomers.
Before going further, let's see how the system works. You can see above a photo of Lagoon Nebula posted on Flickr by reactionwheel (Credit: reactionwheel).
And above is an analysis performed by the Astrometry.net team. The green circles are stars from the index, projected to image coordinates, while the red circles are the original field objects. (Credit: Astrometry.net). You'll find more explanations on this specific page. and here is another link to many other Flickr photos 'solved' by the system.
Now let's read carefully the U of T news release. First, it misspelled the website associated with the project, writing about Astronometry.net and Astonometry.net in their article. The correct site is Astrometry.net and I'm amazed that nobody fixed this at U of T. If I was Sam Roweis, the Associate Professor behind the project, I would not be happy.
Dustin Lang, a computer science PhD candidate who works with Roweis, explains some of the challenges the team found. "We call it a blind astronometry solver. It's a bit like going utside on a dark night and trying to find the constellations, except we're trying to recognize images that come from all kinds of cameras, amateur telescopes, large ground-based telescopes and space telescopes such as the Hubble Space Telescope. Some of the images we are trying to solve cover less than a millionth of the area of the sky -- about 10 per cent of the size of the full moon."
Of course, when you want to achieve such a technical prowess, you need to have some serious computing resources. But exactly how much? When asked what he enjoys most about the project, Lang said, "Working with astronomers is great. They deal with extremely small and extremely large things, so they get to be really good at 'order-of-magnitude' thinking: Is this process going to take a minute, an hour or a week? Do we need 10, 100 or 1,000 computers to solve this problem?"
This project will probably be more helpful for amateur astronomers than professional ones. "Amateur astronomers can take great pictures but they rarely record where their telescopes are pointing -- we can figure out exactly where the image came from and combine images into a high-resolution picture of the sky that is always being updated. Professional astronomers can use this data to look for transient events like comets, supernovae -- things an amateur astronomer may have taken a picture of without even knowing it."
It's time to read the project summary on the Astrometry website. It starts with a bold statement. "We are building an 'astrometry engine' to create correct, standards-compliant astrometric meta data for every useful astronomical image ever taken, past and future, in any state of archival disarray." Wow!!!
And it continues with a description of the search engine. "The astrometry engine will take any image and return the astrometry world coordinate system (WCS) -- a standards-based description of the (usually nonlinear) transformation between image coordinates and sky coordinates -- with absolutely no 'false positives' (but maybe some 'no answers'). It will do its best, even when the input image has no—or totally incorrect—meta-data."
Here is another bold statement about the algorithms which are used by the computer scientists. "We have elucidated and solved a fundamental computer science problem in the field of geometric hashing: the fast and efficient search for matches to patches of a two-dimensional set of points, when the patch to be matched has unknown location, scale, orientation, and completeness or contamination, as well as realistic errors. Efficient and robust algorithms for this matching problem will be the basis for attacking many highly non-trivial problems in pattern matching, data analysis, and computer vision."
For more information, here is a link to a presentation given by Roweis at Google Pittsburgh on the project (PDF format, 49 pages). But for all the technical details, you should visit this page, which is not referenced on the site homepage. And keep in mind that Astrometry is an open source project -- but in beta stage. Still, you can ask the team to download their software -- or send them your sky images for analysis.
Sources: University of Toronto News, June 27, 2007; and various websites
You'll find related stories by following the links below.