Monday, April 12, 2010

Hacking Facial Recognition Software

One thing that I see often in TV is facial recognition software.  For example, in the current season of "24", Chloe O'Brien is able to very easily figure out who someone in based on the digital image of their face.  The way these software programs work is similar (though not as advanced as in "24").  They take a facial image and scan it to determine what the features of the person are, and then compare that to a database of people's characteristics.

Of course, the easy way to beat this would be to wear a ski mask or something, but it would be way too obvious if you were walking around in public with one on.  I often wondered if some sort of plastic surgery would make you harder or impossible to detect, and that certainly is an option.

However, a computer programmer was able to reverse engineer this software to find ways to beat it, in theory by using makeup patterns.  He had three sets of images.  The first set were basic images with no makeup, the second set was images with random patterns, and the third set were images that exploited what he considered potential weaknesses in the facial recognition software.  His conclusions?  The patterns he created all fooled the system, while the random patterns and the blank patterns did not fool the system.

His conclusion?  The images represent potential anti-surveillance makeup.

Now, if someone was walking around with the makeup you see in the images, it might look weird still...but the potential is there.  If someone can wear a weird makeup pattern and throw off these systems - well, these systems need to be more mature and figure these things out.  I am sure the companies will say "well, no one is going to walk around with that makeup" publicly, while privately scrambling to find a fix.

Needless to say, if a graduate student came up with an idea like this...imagine what terrorists who do not want to be found will come up with.

No comments: