Researchers Create Search Tool That Can Identify Nude Pictures

Two computer-science professors have developed a program for spotting pictures of naked people.

The professors -- Margaret M. Fleck and David A. Forsyth -- say they couldn’t care less about identifying pornography. Their research project is simply one step toward a larger goal of developing search tools that can"read” a wide range of pictures.

Dr. Fleck, an associate professor of computer science at the University of Iowa, says searching for nudes presents"a good example, scientifically.” Colors, she says, are the easiest elements for computers to single out. The range of skin colors is limited, even after taking race into account.

The program first looks to see if there are large patches of skin-like colors in an image. It then analyzes the skin-colored areas to determine whether they contain shapes that could be arms or legs.

“It works,” says Dr. Forsyth, an associate professor of computer science at the University of California at Berkeley. When analyzing a set of about 5,000 images, the program correctly found 43 per cent of the nudes. Four per cent of the time, it identified an image as a nude that was not.

What threw the computer off? Dr. Fleck says images of “desserts, certain fruits and vegetables, autumn colors, and cavern scenes” were particularly troublesome. Apparently, earth-toned stalagmites and stalactites can look a lot like arms and legs.