Your face is becoming the key to accessing your money, your devices and could mean the difference between freedom and imprisonment. It's a feature that unlike fingerprints can be scanned at a distance, and it's being used on a massive scale to electronically identify people as they walk past a camera. Here’s how it works, and how it fails.

Facial recognition systems generate what is called a faceprint — a unique code applicable to one individual — by measuring the distance between points like the width of a person's nose.
Tap or click a feature to see what the system measures.

These so-called "nodal points" — there are more than 80 points that a facial recognition system checks — are combined mathematically to build the faceprint, which can then be used to search through an identity database.

The systems used on some personal devices work a little differently. The iPhone X shines a grid of 30,000 infrared dots on a face and makes a crude 3D model. This method works from a metre away.

The Microsoft Hello system takes an infrared picture of a face to allow for traditional facial recognition at a 1m distance. Neither system stores face images, just the resulting faceprint code.

A security camera image can be used to recognise faces from far away. They use a more complex system that looks for faces, re-orients, skews and stretches them. These are then converted to a black-and-white image to make facial features easier for the computer to recognise.
Select a facial recognition stage on the left to see it work.

Facial recognition error rates can be as low as 0.8 per cent, but in practice that would mean eight in 1000 scans could misidentify a murder suspect. A case reported in The Intercept showed how security footage of a bank robber was matched by facial recognition to a man named Steven Talley, who could prove he was not at the bank at the time of the robbery.