Exposing digitally doctored photos
Scientists developing tools to track images, spot manipulation
In the digital age, it is easy to generate and share photos and videos - and just as easy to manipulate them. Visual tweaks can range from small changes, such as cropping photos, to more substantial edits, such as erasing someone from an image or substantially, even unnaturally, enhancing a person’s appearance.
As it has become easier to trick the eyes, a field of forensic imaging has emerged, with scientists creating tools that will be able to test whether and how images have been altered, with plenty of potential real-world applications. News organizations might need to verify whether photos submitted by readers are real. Scientific and medical journals could check whether an image presented as key data carries signs that it has been faked. In a courtroom, such tools could be used to look for signs of tampering in surveillance video or crime scene photos.
Start-up company Fourandsix Technologies, cofounded by Dartmouth College computer scientist Hany Farid, is developing software that will determine whether a photo has been altered since it was first taken - and if so, how.
“We’re in the business of trying to really think like a forger,’’ Farid said.
For example, he tries to think of ways that a forger intent on deleting an object from a photo, or adding one, might leave traces in the image; maybe the way the light falls on the people in the photo from subtly different angles will not be taken into account.
Farid said the company’s first product, software that will be able to tell what kind of camera took a photo and whether it has been altered since it was taken, will be available in March.
The field has taken off, in part because the photos people pass around so easily in saved files have unique digital fingerprints - a distinct signature that is created when a photo is taken.
“You may see a picture - it may be a mountain or a person. What I see is the computational trace,’’ said Ray Liu, a professor of information technology at the University of Maryland, who is also developing techniques to detect that an image has been altered. He added that such technology has become of great interest to the military.
In a very different arena - advertising and magazines - photo manipulation has become increasingly important, because it can create unrealistic visions of body image.
The problem was recognized this summer by the American Medical Association, which officially adopted a policy to discourage advertisers from altering photos in ways that could create unhealthy expectations among vulnerable adolescents. Two years ago, a Ralph Lauren ad came under fire for showing a freakishly thin, digitally altered version of model Filippa Hamilton.
Overuse of airbrushing can also create misleading advertising. In the United Kingdom, an image in an Olay beauty ad featuring a virtually wrinkle-free Twiggy, for example, created a stir in the media and generated complaints from the public two years ago. The image was eventually withdrawn and replaced, when the company found the photo had been retouched around her eyes.
Farid and a graduate student, Eric Kee, last week reported that they had devised a way to quantitatively analyze before and after photos to determine how much digital airbrushing had occurred.
They used measures that could pick up geometric alterations, such as whether a person’s body shape has been changed - their breasts enlarged, waist shrunk, or neck elongated. Another measure quantified changes to attributes such as color, skin tone, or removing blemishes.
The big problem with flagging all photos that have been digitally altered is that many photos with relatively unimportant changes - rotating or cropping an image - might also fall into the same group as those where a waist has been slimmed down or a bust plumped up.
In their paper, published in the journal Proceedings of the National Academy of Sciences, Farid and Kee described their new tool, and demonstrated that their quantitative measures of alterations could pick up on meaningful changes to appearance. The eight measures the Dartmouth team used were, overall, able to score photos’ resemblance to reality similarly to the way human beings would rate them.
“Unlike the blunt instrument of ‘altered’ or ‘not altered,’ it allows you to distinguish - this is really a minor color modification followed by a few modest manipulations, or this person is physically different than the person they started out to be,’’ Farid said.
The researchers proposed that photos be objectively and quantitatively analyzed to rate the amount of manipulation they contain, and that the scores be published alongside photos, providing more honesty.
It’s very easy to get carried away when touching up an image.
“It’s a slippery slope - you can go off the rails pretty quickly,’’ Farid said. He recalled a friend who had touched up a photo for an online dating site that became unrecognizable
“He said, ‘You know it’s a weird thing - I started off with a real picture, somehow I got to the end and it doesn’t look anything like me.’ ’’