Developers use photos fromvarious online sources to
train theirfacial-recognition systems. How do you know if
yourface has been harvested in thisway?You don’tRemove your face from the webCOVER
FEATURE7 - 20August 2019 41and Google Images, to train their
facial-recognition systems. How do
you know if your face has been
harvested in this way? You don’t.How does it identify you?
There are various systems, which use
differen t methods to identify faces.
Essentially, however, they build a 2D or
3D image of your face, then compare
it with other images. Systems that
use 2D imagesrely on common facial
landmarkssuch as eyes, nose and
mouth, including their size, shape andYOUR FACIAL-RECOGNITION QUESTIONS ANSWERED
the distance between them, to produce
a unique faceprint – the facial equivalent
of a fingerprint. However, differen t light
conditions, angles and areas of shadow
can make a face unrecognisable.
This is less of a prob lem with 3D
models, which include additional
data ondepth, curves and so on.How is facial recognitionbeing
usedtoday?
It’s everywhere.Imagine you’re taking a
photo of a friend using an iPhoneXS, for
example. Even before you start, you use
facial recognition to unlock the phone in
the first place. Then, whil e taking the
photo, the camera recognises that
you’re pointing it at a face. Once you’ve
taken it, facial-recognition technology
compares your friend’s face with those
in previous photos, so now when you
tap Search inApple’s Photos app, you
can see every photo your friend appears
in. That photo, alongwith all the
previous ones, is automatically backed
up in cloud storageservices, such as
iCloud and Google Ph otos, which use
more facial recognition to help you
search for people you know there. If
you’ve previously tagged your friend on
Facebook, the social network knows
who they are when you share this new
image there, too.Whatis facial recognition?
There are two basic types of facial
recognition. At its simplest, the
technology recognises that something
is a face. Point your phone’s camera at
one or more people, for example, and if
it detects what look like eyes, a nose
and a mouth in clos e proximity to each
other, it wi ll assume it’s a face and stick
a box aroundit, track it and make sure it
stays in focus. Point your camera at a
crowd and it wi ll identify multiple faces.
More advanced facial-recognition
systems not only spot faces but try to
work out who a particular fa ce belongs
to. When you see someoneyou know,
your brain proc essesa huge amount of
information about them – the size,
shape andcolour of th eir eyes, the
width and leng th of their nose, the size
and shape of th eir mouth, the colour
and style of th eir hair, and so on.
Advanced facial-recognition software
does much the samething.
What is face-recognition
training?
While most people can identify faces
easily, computers find it much more of a
challeng e. Your phone, for example, can
usually pick out faces in a scene, but it
might also tag a tree that has face-like
feat ures in the bark. To make the task
even trickier, faces can be distor ted in
numerous ways– such as when we
smile, fr own, yawn and so on – and they
look differen t depending on the angle.
Feeding lots of photos of th e same face
and others into a computer makes the
recognition models much smarter,
enabling them to identify faces when
only a portion is visi ble, for example.
The developers use photos from all
sorts of sources, including Facebook