Finding Your Place in the Universe
Astrometry.net matches images of the sky to their celestial coordinates, turning anyone into an astronomer
Anna Rothschild • February 19, 2010
Astronomer David Hogg imagines a world where everyone who owns a digital camera would place it on the roof before going to bed, set the shutter to automatic, and point the viewfinder skyward. Then, by streaming the billions of images taken each night onto the Internet, we could all contribute to an ever-changing picture of space, searchable and accessible to anyone with a computer. The resulting archive, says Hogg, would be worth its weight in stardust.
Although Hogg’s dream may seem far-fetched, the New York University professor is already laying the groundwork for it to come true. He is one of the architects of a new system called Astrometry.net, which anyone can use to identify the precise celestial location of an astronomical image. To Hogg, any image that is taken of the night sky is a potential piece of scientific data if its location can be confidently determined.
“My vision would be that all astronomical imaging could be organized in a uniformly searchable way,” said Hogg. “But that system will only work if all of those images are correctly handled, and calibrated, and stored.”
Astrometry.net is a first step towards this goal. Not only can the system identify the celestial location of an image, it can even label objects, like stars and nebulae, in the picture. It is also quite versatile, able to resolve photographs as clear as those taken by NASA’s orbiting Hubble Space Telescope, and as murky as iPhone screen shots. In short, it turns pretty pictures into trustworthy sources of information.
“The nice thing about Astrometry.net is that it allows you to calibrate a wide and varied dataset that is coming from a lot of individuals,” said Cameron Hummels, a graduate student in astronomy at Columbia University who co-directs Columbia’s Astronomy Public Outreach program.
Astrometry.net makes public participation easy. The system is hooked up to Flickr, an online image database managed by users worldwide. So amateurs, even ones with simple digital cameras, can tag their photographs and potentially contribute to science.
Amateur astronomers have historically been important sources of astronomical data. In July 2009, for example, an amateur discovered an Earth-sized scar on Jupiter caused by an asteroid or comet that crashed into the planet’s surface, and in September, another amateur beat the world record for the most supernovae ever spotted by a single person.
“This is one of the cool things about astronomy,” said David Schiminovich, a Columbia University astronomer who uses Astrometry.net. “It’s not the sort of thing where you need to have a particle accelerator. … It’s something that everyone can actually participate in.”
Since the site opened in 2007, more than one thousand people have joined the Flickr group. In the next few years, the Astrometry.net team will mine the calibrated amateur data on the site, looking for unexpected celestial bodies like new supernovae, or potential Earth-colliding asteroids for use in professional analyses.
“Anytime somebody takes a picture of the night sky and puts it on Flickr, there’s a possibility that it contains interesting discoveries,” said Hogg. “So, one of the reasons we’re doing all this is to prepare ourselves for a world in which we could use the whole internet as if it was a telescope.”
Hobbyists look forward to a day when they could easily contribute to professional astronomy. “That would be fantastic. I’d love it if my pictures got in there!” said David Thompson, an amateur astrophotographer from the United Kingdom who uses the system to label his images in Flickr.
Astronomy.net doesn’t just resolve amateur images. The program was originally designed as a way to automate the image calibrations that cost the professional astronomical community time and money each year.
The creators turned to a “simple, fairly old-school kind of approach in computer vision” explained Dustin Lang, the University of Toronto computer scientist who programmed Astrometry.net. This approach is known as geometric hashing, a way to identify objects in an image using their shape, without knowledge of their scale or rotation.
The program picks four stars in a novel image, and builds a geometric shape with them. It then matches that shape to a reference catalogue of around 800 million sets of four stars found in the sky. When it finds a similar shape in the database, the system makes a prediction about the coordinates of the image, and uses other stars in the picture to test that prediction. It repeats this process until it is sure it has found a piece of sky that matches the image.
Schiminovich said that he would consider using the program on rockets to replace the current in-flight software that determines where the rocket is pointing. While it is unlikely that NASA will be using the program on expensive missions any time soon, Schiminovich believes that Astrometry.net “is almost guaranteed to be more reliable and faster than what you’d spend a lot more money on.”
Speed is a virtue if you want to create a model of space that includes practically every image ever taken, since you will have to calibrate, or recalibrate, a vast amount of data. New York University alone has about 40 terabytes of astronomical data — that’s the same amount of data contained in four copies of the entire printed collection of the U.S. Library of Congress.
A lot of the existing data has been mislabeled or corrupted, due to human errors or even faulty software. But according to Schiminovich, those errors could be corrected by running them through the system.
“There’s a whole historical record of the sky that is lost,” he said. Astrometry.net “allows you to go back into the record to recover this stuff.”
So far, no one has created a system that reliably catalogues all forms of image data, from large telescopic photographs, to historical plates, to amateur snapshots. But there are sites that are beginning to make space “searchable.” For example, both NASA’s SkyView Virtual Observatory and Google Sky allow anyone to find image data from a number of professional space surveys by simply typing in some celestial coordinates.
But Hogg has dreams of, well, astronomical proportions.
“Ideally we would be assembling everything into a continuous, changing, updating model of the sky,” said Hogg. “So then you would really have assembled, all in one place, all the knowledge that there is about the sky.”
I sounds fantastic, but what about those who are retouching images and insert, say, a fake supernova? Or an accidental reflection in the lens? Lots of variables that I am sure you already though.