Facial recognition as a form of authentication, like
many of the technologies that have bubbled into
reality in the past few years, is something that Sci-
Fi has been promising for decades. Alas, it’s… not
always very good. Android’s face-based unlock
mode, for example, tends to be fun for a few
minutes before it just gets frustrating.
IsItYou, a company debuting at Disrupt SF 2014
today, claims to do it much, much better.
Presenting on stage today, company co-founder
Benjamin Levy claimed that their algorithms have at
least two advantages over Google and the rest: it
works in extreme lighting conditions (like a dark
room, where the only light is from your screen), and
its anti-spoofing tools make tricking the algorithm
tough.
I spoke to Levy, who explained why better facial
recognition is so crucial. “On most of the phones
out there today, the facial recognition fails 2 or 3
times out of ten. With our solution, the false-
negative rate is closer to 2 or 3 out of 10,000″
False negatives are one thing — but what about
false positives ?
Android’s facial recognition got some bad marks
early on when someone managed to trick it using
nothing but a printed still photo. Google later fixed
this by requiring the user to blink on command —
but that solution feels like training the user, rather
than training the phone.
IsItYou analyzes not just your face, but how it
expects the lighting on your face to behave in
different situations. Light bouncing off of a flat
surface (like a printed photo, or a tablet screen)
behaves much differently than light bouncing off
your face. They demonstrated their algorithm
rejecting a printed photo of Levy’s face — then,
taking it one step further, had someone don a 3d
printed mask of Levy’s face. It, too, was rejected
As for the business model: IsItYou intends to be a
solution for others to plugin to their apps, whether
it’s your bank or your favorite social network.
They’d charge per authentication.
The company plans to launch into private beta in Q4.
source: techcrunch.com