Digital detectives discern Photoshop fakery
New software combs for clues in al Qaeda tapes, Harry Potter pages, and celebrity waistlines.
Subscribe Today to the Monitor
Earlier this month, Paris Match ran this picture of the shirtless Mr. Sarkozy sans poignees d'amour – or love handles. With a slight digital nip-tuck, the magazine trimmed the flab that peeked above the presidential waistline.
"The position of the boat exaggerated this protuberance," explained Paris Match last week, after another French weekly, L'Express, exposed the touch up. "The correction was exaggerated during the printing process."
As image-manipulation software becomes easier to use and harder to detect, the problem of tampering has spread far beyond such celebrity "corrections." While fudged paparazzi moments do little more than embarrass editors, there are far more important – and sometimes illegal – fakes to catch.
"The most common examples of doctored photos occur in the media, but there are serious cases of image manipulations in security and investigations as well," says Cynthia Baron, author of the book "Photoshop Forensics," scheduled for release in December. "There are researchers working as the frontline of defense against digital fraud." And they're developing some very tricky ways to spot shams.
Over the past six years, computer science professor Hany Farid has become something of a digital detective. While Paris Match's virtual liposuction was exposed because the unaltered photo ran in several other publications that week (including the Aug. 6 Monitor), Mr. Farid doesn't need the original to reveal tampering.
"There's no way to push a button and tell if it's real, but there are tests we can run that allow us to be pretty sure if it's a fake," says Farid.
Some of the investigative techniques are simply teaching a computer to spot what the untrained eye is too lazy to see. If a figure from one photo has been edited into another, there are almost always imperfections – subtle inconsistencies in the physics and geometry of the combined image. The vanishing points might be off, or the shadows cast from two or more objects may contradict one another.
"These are things humans are really bad at noticing," says Farid. But to a computer, the subtle differences are obvious.
Farid can now run possible forgeries through a gamut of tests, even checking the light reflections in people's eyes to triangulate the location of the flash camera that took the picture. If the analysis of subjects in a photo shows that the camera had to be in multiple places at once, the shot's a fake.
Courts face off with digital fraudsters