Almost everyone has scratched their head once or twice, thinking they’ve “seen that face before”. In some respects, that’s probably the same momentary technological thought a facial recognition system might have.
However, this type of artificial intelligence is supposed to instantly jump past any temporary brain block. But what if facial recognition systems were vulnerable to being tricked? With an increasing emphasis on facial recognition, logically, someone would test the system.
Researchers from the Blavatnik School of Computer Science and the School of Electrical Engineering in Tel Aviv have done just that. The theory they applied worked. What the researchers did was create what they called nine “master key” faces.
When they tested these “master faces”, researchers were able to impersonate nearly half the actual faces in a dataset. The experiment tested three of the leading face recognition systems. The master faces could impersonate over 40 percent of the population.
There wasn’t any need for additional information or data about the person they were attempting to trick the system into identifying. These same researchers used their testing criteria to create a block of “master fingerprints” as well.
In the facial recognition experiment, researchers tested Dlib, FaceNet, and SphereFace. They used various algorithms and computer-generated iterations to narrow down sophisticated facial features. This experiment wasn’t focused on how to trick facial recognition systems.
It was trying to improve them. Nevertheless, if anyone thinks that a less noble group of geek-types might try the reverse, it is rather technologically naive. If the good guys can figure out how to trick facial recognition, you can bet the not-so-good guys are working feverishly to do the same.
What this does prove, beyond a reasonable doubt, is that facial recognition needs to go back to the proverbial drawing board. These systems have already failed many times. Law enforcement, attempting to use this flawed technology, has made multiple wrongful arrests.
As promising as some feel facial recognition might become, it appears to be easily tricked. For the time being, old-fashioned fingerprints will have to continue as a primary means of uniquely identifying people. But don’t be surprised if “the eyes” have it in the not-so distant future.
The post Experiment Creates a set of “Master Faces” to Trick Facial Recognition Systems appeared first on Steadfast Clash.
Source by [author_name]
