Well-meaning researchers from Carnegie Mellon University destroyed the credibility of all photos on the web today. Alexiei Efros, assistant professor of computer science and robotics lead the team that created two related systems that, together, will do what governments and corporations could only dream of (until now).
The first program, called Photo Clip Art, uses thousands of labeled images from a site called LabelMe. These images can be added to photos. The system analyzes the geometric context of a photo, and place objects within the scene, adjusting its size as necessary to put it in proportion to other objects of equal distance from the camera.
The second program is called Scene Completion. It draws upon millions of photos from the Flickr website to fill in holes in photos. These "holes" are created when editors decide that some part of a picture is unsightly, and wants to remove it.
(Scene completion work flow)
Clockwise from top left 1) the original photo, 2) undesirable material removed, 3) thousands of possible matches scanned and 4) the new, doctored photo
The Associated Press published the photo two years after it was taken. The image was snapped in August, a time when polar icecaps naturally melt and the wider shot would have shown that the bears were near land mass. Oh yeah, and polar bears can swim! All this didn't stop Al Gore from using the photo as emotional propaganda to support his case.
Using photos of oft-snapped subjects (like Notre Dame) scraped from around the Web, Photosynth (based on Seadragon technology) creates breathtaking multidimensional spaces with zoom and navigation features that outstrip all expectation. Its architect, Blaise Aguera y Arcas, shows it off in this standing-ovation demo. Curious about that speck in corner? Dive into a freefall and watch as the speck becomes a gargoyle. With an unpleasant grimace. And an ant-sized chip in its lower left molar. "Perhaps the most amazing demo I've seen this year," wrote Ethan Zuckerman, after TED2007. Indeed, Photosynth might utterly transform the way we manipulate and experience digital images.