Summary

  • A July 2007 project announcement invited Flickr users to submit sky photos for automatic calibration; object annotations followed in July 2008.
  • The familiar upload concealed demanding work: propose a position from star patterns, check it, and cope with defects in both photographs and reference catalogues.

A photograph of the night sky can contain an impressive number of stars and remarkably little information about where to look. Its owner may know exactly what happened. The next person gets a dark rectangle and a confidence problem.

Astrometry.net found a pleasantly ordinary place to deal with that problem: a Flickr group. Its 29 July 2007 announcement described automatic processing through the project's blind solver and Flickr's API. Any Flickr user could submit. A photo-sharing habit had acquired an astronomical back office.

The answer came back to the picture

The route into the service mattered. Christopher Stumm, who proposed the Flickr connection, was already a user. In a February 2009 interview on Flickr's developer blog, he pointed to familiar uploads, group submission, image annotations and an API that made the connection easy.

There was a concrete reward for contributing. The project announced object annotations on 6 July 2008: newly submitted, solved images could acquire labels for major objects. The photographer could see what the machine had found on the photograph itself. A celestial address becomes considerably more engaging when it points out something in your picture.

The incoming material had personality. Stumm recalled backyard photographs, hand-drawn attempts to fool the system and a screenshot from an iPhone sky application. The Internet had discovered a scientific instrument and immediately checked whether it could take a joke.

Four stars make a suggestion

The service's 2009 paper explains the serious work inside. It detects compact sources, builds geometric patterns—typically from four stars—and searches a prepared index for similar arrangements. A candidate match proposes the image's pointing, scale and orientation. Other detected stars then help test that proposal.

That last step is where resemblance has to earn its living. A suggestive shape alone could send a photograph to the wrong patch of sky.

The authors published a wonderfully awkward example. An International Space Station trail produced a line of apparent sources. Writing on a scanned reference plate had also produced false sources. The system aligned the two. One piece of human activity in the sky had met another at the edge of a photographic plate.

A useful reply beats a mysterious upload

The Flickr connection joined two kinds of effort. Photographers supplied images through a service they knew; the astronomy project supplied a way to locate them and return visible information. Stumm described amateur data that could be excellent yet difficult to find or use because useful metadata was missing.

JPEG support, recorded in February 2007, had already brought ordinary web images within reach. The group supplied an entrance; annotations supplied a reason to inspect the answer. Behind that friendly exchange sat reference catalogues, detection choices and people responsible for keeping both useful. The photograph had found an address. Somebody still had to maintain the address book.