Researchers discover alghorithmic method of identifying naked bodies

It is a disturbing fact that 28 percent of men admit to viewing pornography at work (the survey hopefully did not actually include pornographers, who at least have an excuse). Such behavior leaves their employers open to charges of sexual harassment at worst, and reduced efficiency at best. Many companies simply limit the Web sites their employees can visit using various commercial filters, but many resist this approach because such software can prevent access to legitimate sites while at the same time creating a climate of distrust that can upset workplace morale. Two researchers have recently been awarded a patent for an algorithm that overcomes all these problems: instead of identifying pornographic Web sites by textual tabs (which can mistakenly limit access innocent pages containing the word “breast”, or even academic research on pornography), the algorithm instead detects the presense of naked flesh on a page.

Justice Potter Stewart notably remarked about pornography that “I know it when I see it.” So too does the program developed by Margaret Fleck and David Forsyth, which tries to overcome the problem that pornographic images may have varied color backgrounds and depict multiple figures or partial figures in an assortment of positions that the reader can only imagine. The images, too, are taken from a wide variety of camera angles and under a wide range of lighting conditions. The Fleck-Forsyth algorithm addresses the challenge by first locating images containing large areas whose color and texture is appropriate for skin, and then groups elongated regions into possible human limbs and connected groups of limbs. Images containing sufficiently large skin-colored groups of possible limbs are reported as potentially containing naked people.

We note this story for two reasons. First, because sexual harassment in the workplace is a serious concern for IT planners, and second because the algorithm demonstrates an innovative approach to identifying and categorizing data. We often report on video analytic software that is able to identify suspicious behavior, and we wonder if the Fleck-Forsyth algorithm may have appropriate uses in that sector. Not to identify naked people — or, rather, to identify people as being naked — but as a method of quickly establishing common features in order to quickly sort and categorize images. Perhaps the system could even be used in the exact opposite manner than Fleck and Forsyth imagine: at a beach resort, all are expected to be in various stages of undress. Quickly identifying those wearing full-length coats and heavy dungarees might help stop terrorists in their tracks.

-read more in this research summary; see also this patent award