Make like a Dandy Highwayman to beat Face Recognition Software

Spoofing biometrics has become a mini-industry, as one would expect as the technologies of recognition become more pervasive. And not all of these methods are high-tech. Tsutomu Matsumoto’s low tech ‘gummy fingerprint‘ approach to beating fingerprint recognition is already quite well-known, for example. I’ve also seen him demonstrate very effective iris scan spoofing using cardboard irises.

Facial recognition would seem the most obvious target for such spoofing given that it is likely to be the system most used in public or other open spaces. And one of the most ingenious systems I have seen recently involves a few very simple tips. Inspired by the increasing hostility of legal systems to masks and head coverings, CV Dazzle claims to be an ‘open-source’ camouflage system for defeating computer vision.

Among the interesting findings of the project, which started as part of the Interactive Telecommunications Program at NYU, is that the more complex and high-fashion disguise-type attempts to beat facial recognition did not work as well as the simpler flat camouflage approaches. The solution suggested thus involves many of the same principles as earlier forms of camouflage: breaking up surface patterns and disguising surface topography. It uses startling make-up techniques which look a bit like 80s New Romantic face painting as deployed by Adam and the Ants – hence the title of this post! The system concentrates especially on key areas of the face which are essential to most facial recognition software systems such as the area around the bridge of the nose, cheekbones and eye socket depth.

Results from the CV Dazzle project

So, will we see a revival of the Dandy Highwayman look as a strategy of counter-surveillance? Or more likely, will social embarrassment and the desire to seem ‘normal’ mean that video surveillance operators have a relatively easy life?

Adam Ant in the early 80s