The application — created by Hany Farid and Eric Kee of Dartmouth College in Hanover, New Hampshire — uses eight statistical parameters to detect real changes on the model, not general modifications like cropping or colour adjustments. It can sniff geometric modifications like reshaped face features or body parts used to make models appear thinner. It will also detect texture changes, like smoothed skin.
They trained their program using input from 350 volunteers, obtaining an 80 per cent accuracy. The first version of the software is tuned for faces, but the scientist claim that they can easily repurpose it to detect manipulations in any kind of images, such as those presented by scientists in research. Or, you know, the classic Apple rumour spy shots.
For now, they hope that advertisers would include the manipulation percentage next to the model faces and bodies, as a warning label to women who may fall into self-confidence problems that eventually may lead to anorexia and other eating disorders. That’s a great idea that is not coming soon enough. [Nature]