This AI Tries To Figure Out If You're A Real Person Posted on : Aug 23 - 2017

Earlier this year, Samsung launched its premium Galaxy S8 phone with the ability to unlock the device simply by placing your face in front of the camera, but it didn't take long for someone to figure out how to trick it. Mere days after the phone came out, a user showed how he could take a selfie on his own phone and then unlock the Galaxy S8 trained on his face by showing it the selfie. That's hardly a feature worth trusting.

A small startup, called FaceTec, said it has created software that can detect whether the face in front of the camera is actually a real person (and not just a picture or a video) using a collection of machine learning techniques.

FaceTec's authentication process requires the user to hold the phone about one foot away from their face and then pull it in to about 8 inches. The software takes the images from the two positions and recreates a 3-dimensional model of the person’s face. Each frame of video is then analyzed using around 30 machine learning classifiers that look at various aspects, such as skin texture, the ways shadows change, and the reflection of the phone screen in the eyes.

FaceTec isn't using any deep learning techniques, a currently popular brand of machine learning that's making huge strides in areas like image and speech recognition. The company has tested deep neural networks in its data center with the help of Google's deep learning framework, TensorFlow, but said it doesn't need the techniques to achieve accurate results right now. Also, FaceTec's artificial intelligence runs completely on the phone, so it needs to run as efficiently as possible, as well as take up as little space as possible. View More