Adobe Tackles Photo Forgeries
Wednesday, December 26th, 2007
A suite of photo-authentication tools under development by Adobe Systems could make it possible to match a digital photo to the camera that shot it, and to detect some improper manipulation of images, Wired News has learned.
Adobe plans to start rolling out the technology in a number of photo-authentication plug-ins for its Photoshop product beginning as early as 2008. The company is working with a leading digital forgery specialist at Dartmouth College, who met with the Associated Press last month.
The push follows a media scandal over a doctored war photograph published by Reuters last year. The news agency has since announced that it’s working with both Adobe and Canon to come up with ways to prevent a recurrence of the incident.
“Fundamentally, our values as a company requires us to build tools to detect tampering, not just create tampering,” said Dave Story, vice president of product engineering at Adobe.
Photo manipulation is nothing new. During the Stalin era, Soviet officials frequently vanished from official photographs after falling out of favor at the Kremlin.
But the advent of Photoshop and its variety of tools has made it easier for photographers to tinker with images after they’re captured. By the same token, the internet has allowed skeptical bloggers around the world to analyze photos in depth, and expose chicanery.
In the most famous recent case, a blogger uncovered the doctoring of a war photo taken in Lebanon by Reuters photographer Adnan Hajj. The photographer was fired, and Reuters has since clarified its rules about the use of Photoshop.
AP has not had a similar scandal but is still on guard. “When we look at the manipulated images that we have come across historically in the AP, it’s a tiny, tiny percentage. But all it takes is one or two and the effects are huge,” said Santiago Lyon, director of photography for the AP, which handles about 750,000 photographs a year.
Despite the potential for disaster, photo editors still rely on their own eyes to detect forgery, even as advances in Photoshop technology make manipulations even less obvious. “We do really advanced math so you can’t detect what’s going on, and we’re getting better at that every year,” Adobe’s Story said.
In a speech in Tel Aviv in December and a blog entry, Reuters CEO Tom Glocer said his company is working with Adobe and Canon to create an “audit trail” that would reveal changes made to an image. Neither Reuters nor Canon would provide details on the plan.
Officials at the AP, meanwhile, met Feb. 5 with Hany Farid, a Dartmouth College professor who studies ways to detect digital forgery, Farid said.
Farid is working with Adobe on its upcoming photo-authentication plug-ins, which will rely on mathematical algorithms to pinpoint signs of manipulation.
Among other things, Adobe is developing a tool that will detect the use of the copying tool known as the “clone stamp.” The tool will identify when two areas in a photo are “impossibly similar,” Story said.
Adobe expects another tool will perform an analysis similar to firearm ballistics — confirming the model of camera that took an image, and matching the image to the individual camera, if it’s available.
The company also hopes to develop a plug-in that will detect if a photo has been changed at all since it was taken. According to Farid, this is possible because cameras don’t record all the pixels needed for a color image, but instead estimate some colors through a process known as color reconstruction, or demosaicing.
A camera’s demosaicing process creates connections between pixels, and “when an image is re-touched, it is likely that these correlations will be destroyed. As such, the presence or lack of these correlations can be used to authenticate an image, or expose it as a forgery,” Farid writes in an explanation (.pdf) of the technology he is developing.
Lyon said AP might ultimately apply manipulation-detection software to photos from “casual freelancers” or handouts from government agencies, entertainment organizations and military officials.
The challenge, Farid said, is to figure out how to detect inappropriate manipulations and ignore ones that are allowed in media photographs, such as cropping and color enhancement. “We can’t say, practically, that you can’t do anything to the image,” he said.
Wrong results — false positives, in particular — appear to be the Achilles’ heel of photo authentication technology. The software is “statistical in nature, and there are a lot of assumptions involved,” said Nasir Memon, a professor of computer science at Polytechnic University who studies digital forensics.
“You always have false positives,” Memon said. “Even if you’re 90 percent accurate … you’ll be telling 10 percent of the people that their image is fake when it’s not.”
Story said Adobe is aware of the potential risk of false positives and will continue trying to perfect the technology for the next one to three years before releasing the plug-ins. “We want to get them to a stable enough place and have enough understanding of how to use them properly that you won’t come to invalid conclusions.”
Tags: amp, cameras, canon, ceo, colors, Digital, Government, images, particular, photo, Photograph, photographer, photographers, photographs, photography, Published, ter