A pair of researchers from Dartmouth University have devised a metric for quantifying how much a digital photograph has been altered by digital photo-editing techniques.
Such a rating could be placed alongside digitally retouched photos in the hopes that the viewers would understand that the images have been altered. "This is about informing the public," said Hany Farid, the Dartmouth computer scientist specializing in digital forensics and image analysis who led the research. "A rating would balance the rights of the publishers and advertisers with the public's right to know."
Digital retouching has been a source of growing controversy over the past few years as magazine publishers and entertainment media websites have altered photos to make them more appealing, using software such as Adobe Photoshop. A website set up by Farid illustrates how the publishing industry manipulates images, showing in one case how a model's breast size was enhanced and, in another, how wrinkles were removed from actor George Clooney's face.
A growing body of scientific literature suggests that altered images can subtly distort people's perceptions, encouraging them to grow dissatisfied with their own bodies and even to spur eating disorders. Digital retouching may also run afoul of laws that guide truthfulness in advertising in some countries, such as the United Kingdom. As a result of these concerns, legislators in the U.K., France, Norway and other countries have proposed mandatory labeling of altered images.
Farid, along with graduate student Eric Kee, developed a single metric, ranging from one to five, that summarizes how much a given image has been altered. A score of one represents minimal retouching, while five represents a very radical reworking of a photograph. They published details of how they generated this metric in a paper, "A perceptual metric for photo retouching," which appears in the current issue of the Proceedings of the National Academy of Sciences.
The metric is a composite of eight different summary statistics that take into account different aspects of geometric distortion, in which the image of a person's body is reshaped through digital means, and photometric distortion, in which the skin or some other attribute is smoothed over or altered in some other way.
To establish a "perceptually meaningful metric," the researchers asked 350 human observers to rank 450 original and retouched photos by the degree to which they felt were manipulated, on a range from one to five, with five being the most pronounced.
Part of the purpose behind development of these metrics is to offer a more nuanced summary of how a photo has been retouched, Farid said. Much of the proposed labeling legislation specifies that if a photo was altered at all, it should be labeled. "A simple label was too blunt of an instrument," Farid said. Mild retouching may only consist of color correction, or stray hairs being removed, while a more severe alteration would consist of removing all facial wrinkles or making someone look thinner by some large degree.
"It would be nice to distinguish between significant and insignificant photo alterations," Farid said.
The metrics could be used in a number of ways. The metrics could be used as part of a plug-in for photo-editing software, in which the software would measure the changes and display the score to the photo editor. It could also be used to support any labeling laws that do get passed: An image published in a magazine or posted online would be accompanied with its score, as a way to alert viewers.
In addition to its use for advertising and publishing, the metric could also be used to gauge other digital manipulations as well. Photography competitions could use it as a guideline for allowable changes in their entries. Scientific journals could use the technology to weed out image fraud, a growing concern in the academic community.
Adobe, Microsoft and the National Science Foundation all funded this work.
Join the CIO Australia group on LinkedIn. The group is open to CIOs, IT Directors, COOs, CTOs and senior IT managers.