Photoshopped or Not? A Tool to Tell

From the NYTimes, read the original article By STEVE LOHR, Published: November 28, 2011

“Fix one thing, then another and pretty soon you end up with Barbie,” said Hany Farid, a professor of computer science and a digital forensics expert at Dartmouth.

And that is a problem, feminist legislators in France, Britain and Norway say, and they want digitally altered photos to be labeled. In June, the American Medical Association adopted a policy on body image and advertising that urged advertisers and others to “discourage the altering of photographs in a manner that could promote unrealistic expectations of appropriate body image.”

Dr. Farid said he became intrigued by the problem after reading about the photo-labeling proposals in Europe. Categorizing photos as either altered or not altered seemed too blunt an approach, he said.

Dr. Farid and Eric Kee, a Ph.D. student in computer science at Dartmouth, are proposing a software tool for measuring how much fashion and beauty photos have been altered, a 1-to-5 scale that distinguishes the infinitesimal from the fantastic. Their research is being published this week in a scholarly journal, The Proceedings of the National Academy of Sciences.

Their work is intended as a technological step to address concerns about the prevalence of highly idealized and digitally edited images in advertising and fashion magazines. Such images, research suggests, contribute to eating disorders and anxiety about body types, especially among young women.

The Dartmouth research, said Seth Matlins, a former talent agent and marketing executive, could be “hugely important” as a tool for objectively measuring the degree to which photos have been altered. He and his wife, Eva Matlins, the founders of a women’s online magazine, Off Our Chests, are trying to gain support for legislation in America. Their proposal, the Self-Esteem Act, would require photos that have been “meaningfully changed” to be labeled.

“We’re just after truth in advertising and transparency,” Mr. Matlins said. “We’re not trying to demonize Photoshop or prevent creative people from using it. But if a person’s image is drastically altered, there should be a reminder that what you’re seeing is about as true as what you saw in ‘Avatar,’ ” the science-fiction movie with computer-generated actors and visual effects.

The algorithm developed by Dr. Farid and Mr. Kee statistically measures how much the image of a person’s face and body has been altered. Many of the before-and-after photos for their research were plucked from the Web sites of professional photo retouchers, promoting their skills.

The algorithm is meant to mimic human perceptions. To do that, hundreds of people were recruited online to compare sets of before-and-after images and to determine the 1-to-5 scale, from minimally altered to starkly changed. The human rankings were used to train the software.

His tool, Dr. Farid said, would ideally be a vehicle for self-regulation. Information and disclosure, he said, should create incentives that reduce retouching. “Models, for example, might well say, ‘I don’t want to be a 5. I want to be a 1,’ ” he said.

Yet even without the prod of a new software tool, there is a trend toward Photoshop restraint, said Lesley Jane Seymour, editor in chief of More, a magazine for women over 40.

Women’s magazine surveys, said Ms. Seymour, a former editor of Marie Claire and Redbook, show that their readers want celebrities to “look great but real.”

“What’s terrific is that we’re having this discussion,” she said. But readers, she added, have become increasingly sophisticated in understanding that photo retouching is widespread, and the overzealous digital transformations become notorious, with the before-and-after images posted online and ridiculed.

“Readers aren’t fooled if you really sculpt the images,” Ms. Seymour said. “If you’re a good editor, you don’t go too far these days. If you give someone a face-lift,” she said, adding, “you’re a fool.”