Feb. 19, 2007 at 9:55 PM ET
The digital world has made it embarrassingly easy to fake a photograph, as illustrated by the doctored Beirut-bombing photo, the bogus Kerry-and-Fonda pairing and the Katie Couric slim-down plan. The good news is that digital tools are providing new ways to detect image manipulation - in fact, a Dartmouth College researcher says he's helping The Associated Press develop a system for finding fakes on the fly. The bad news is that this could mark just one more phase in an "arms race" between the fakers and the fake-fighters.
Farid is hoping that his approach to "camera ballistics" will take some of the controversy out of the news-photo business. Just as a handgun leaves a signature on bullets for ballistics experts to analyze, digital cameras leave a signature on the pictures that they record - in the subtle pattern of electronic noise distributed through the data, as well as in the algorithms used to compress the image for storage.
The system that AP is considering would give a quick scan to the thousands of photos coming in from shooters around the world, Farid said. If there's something inconsistent about a picture's digital signature, that could raise a red flag.
Farid hopes such a system could be developed not only by AP, but by the world's other major photo agencies as well. "Just telling people that 'we have a policy and we're looking' will scare off most people," he said.
Digital analysis should be able to pick up your typical photo fakery, but Farid is under no illusion that computer science can bring a complete end to doctored images.
"It is an arms race, and I can tell you who's going to win: It's not us," he said. "It's the same as spam vs. anti-spam filters."
Farid acknowledged that the payoffs for doctored pictures might not be as obvious as they are for, say, sending massive amounts of spam or counterfeiting $100 bills. "Although I would argue that photographers, especially free-lancers, do have a financial interest," he added. "To get their images published, there is a huge demand for that super-sexy, dramatic photograph."
Speaking of payoffs, millions of dollars can be at stake when it comes to identifying fakes in the art world. For instance, is that "Jackson Pollock" actually a Jackson Pollock? That question has generated a lot of controversy in the past month: Dartmouth researcher Daniel Rockmore said digital fractal analysis has played a supporting role in excluding some works attributed to Pollock.
Ellen Handy, an art historian at the City College of New York, said the Pollock case is far from settled: "Opinions in our art historical community are equally divided."
At the same time, she welcomed anything that can help historians understand the scientific factors behind artistic genius. "We urgently need all the tools we can get," she said.
Rockmore, an expert in the emerging field of digital stylometry, said computers provide just one line of evidence for art sleuths. "Asking for this machine that's going to say 'fake or not fake' is too much," he acknowledged.
But the technique has already turned up some interesting results. Farid, Rockmore and their colleagues at Dartmouth have used computer analysis to determine that parts of Perugino's "Madonna With Child" were done by the master's apprentices. They've also found geometric characteristics, called wavelets, that can distinguish the works of Pieter Bruegel the Elder from imitations (PDF file).
The technique is controversial today, but Rockmore hopes that digital stylometry will eventually become a universally accepted tool of the art curator's trade. It wouldn't be the first time. "Fifty years ago," he said, "no curator wanted to look at an X-ray of a painting."