Friday, November 2

Searching for the perfect image

Belgian artist Martijn Hendriks Untitled I (Google Sleep) consists of 10 Lambda prints which he handselected from an archive of 1000 images found through Google Image search. But why stop there?

The rise of broadband, cheap digital cameras, and social networks like Flickr, PhotoBucket, and Fotki built around tagging and sharing photos has meant there are millions of shots publicly available. And while stock imagery sites like Getty keep redesigning to make finding images easier, scientists like James Hays and Alexei Efros at Carnegie Mellon University employ a new 'brute strength' visual approach.

Their paper, recently presented at Siggraph, describes how algorhythms can power their way through a database of millions of shots, returning a piece (actually lots of pieces) that could match the missing section of an image. Don't like those rooftops in the foreground? Highlight it and choose from alternatives that match the perspective, colour, and lighting in your photograph: trees, a calm bay scene, or a flotilla of ships blended into the original.

It's a small world sharing a large amount of data. In a test case, Hays selected some scaffolding on a European city monument he didn't like. Combing through the millions of possibles, the result came back. The same shot. From the same angle. Taken just prior to construction by another tourist.

No comments: