Cheating face detection the Twisted Sister way

March 15, 2011 by Dave Haynes

I have been doing some writing lately about face detection technology and certainly getting better educated about the issues and arguments around privacy issues as it relates to cameras and software looking for and bracketing faces.

Engadget has an interesting piece up about an NYU student who figured out how face detection could be thwarted by wearing makeup and hair that would have earned you a spot in the background of a Twisted Sister video.

… Adam Harvey, a student in NYU’s Interactive Telecommunication Program, has discovered that some over the top face makeup applied in just the right way can thwart most facial recognition software. Dubbed CV Dazzle (after the Dazzle camouflage used in World War I), the makeup works simply by enhancing areas of the face that you otherwise wouldn’t ordinarily enhance — so instead of applying the makeup around your eyes, you’d apply some on your cheeks and effectively “invert” that area. According to Harvey, that method is effective at blocking the face recognition used by Facebook, Picasa and Flickr — and it doesn’t simply cause some mild confusion, it actually prevents the software from detecting any face at all.

The technology that Intel and similar companies use for audience counting is face detection, and works by looking for patterns of what the machine and software understand to be a face. So it is similar, but undoubtedly also quite different.

So … for people really concerned about their privacy as they shop, they can IN THEORY beat the system by going out in glam makeup. The camera may not count then, but of course everyone else in the place will be staring at them.

 

 

Leave a comment