Faces of two women — original photos at left, Beautified version at right

Ah, digital photography — so full of promise and so full of mischief! Now, in addition to software that slims us down to make us look more attractive and athletic in photos, there’s software that operates at a much more subtle level to “beautify” our faces, using mathematical formulas to morph the original face into a version that’s very similar but (theoretically) more attractive. At the same time, the inventors, Tommer Leyvand and three colleagues at universities in Tel Aviv, say the improved version always shows what they call an “unmistakable similarity” to the original.

The screenshot showing the Digital Face Beautification software in action has the original photo on the left and the beautified version in the center. At the right is the Image Warp, using “digital graph paper” to show how the program distorted — oops, beautified! — the subject’s features. The jawline has been broadened to make the face more rounded; the lips have been moved up a bit and the upper lip made fuller; and the nose has been moved up and shortened. The net result, to my eyes at least, is a picture that looks like the subject’s slightly more attractive sister.

Ironically, though no doubt the underlying mathematics are sophisticated, the work to date is ultimately based on the opinions of only a few people, 68 men and women from Israel and Germany, aged 25 to 40, who picked the most attractive faces from a large number of “head shots” of white males and females. (The researchers have not yet created a program for nonwhite racial and ethnic faces.)

The programmers took the “most attractive faces” data and used an algorithm that related 234 measurements between key facial features — the distances between lips and chin, between the forehead and the eyes, and between the eyes themselves, for instance. This enabled them to program a computer to analyze these distances in any face photograph and to generate the “ideal” face most similar to the original face.

Leyvand et al presented their research at the annual SIGGRAPH computer graphics conference. Although they call their program Digital Face Beautification, it deals only with the shape of the face and its features — there’s no digital retouching to smooth out wrinkles and eliminate blemishes, nor to modify hair or eye color.

As The New York Times points out, psychologists, philosophers, and feminists say this program, like other attempts to use objective principles or mathematics to define beauty, raises many questions about what constitutes “ideal” beauty. “Does a supposedly scientific definition merely reflect the ideal of the moment, built from the images of pop culture and the news media? ‘How can they prove it?’ said Lois W Banner, a historian who has studied changing beauty standards, referring to scientific efforts to define attractiveness… ‘They are never going to get away from the cultural influence.’”

For the Times article, a 25-year-old woman agreed to be photographed and to have her image beautified by the computer program. “She said she was struck by how different she looked in the second shot. ‘I think the after picture looks great, but it doesn’t really look like me at all,’ she said in an e-mail message. ‘My entire bone structure, face shape and eye size is different, and my lip color looks changed as well.’ She added, ‘I would like to keep my original face.’”

Brigitte Bardot might agree. When a glamour photo of her as a young actress was run through the Digital Face Beautification program, “her full and puckered lips were deflated, and the world-famous beauty seemed less striking — less like herself,” said the Times.

Leyvand says there may be practical applications for his software in advertising, film-making and animation. And he has been contacted by interested plastic surgeons. But if Beautification gets built into digital cameras some day, the way Face Recognition and Smile Recognition already have been, it will mean digital photography is taking one more step away from reality.

(No Ratings Yet)
 Loading ...