How does face unlock decide it is you?
Face unlock decides whether the person looking at the phone is its owner. It compares a fresh camera picture with a saved picture of the owner's face and turns the comparison into a match score. If the score is at or above a line, the phone unlocks. No line is perfect, so it can make two kinds of mistakes.
In one line: A program that compares the face in front of the camera with the owner's saved face and unlocks only if they match closely enough.
Last reviewed: 25 October 2026 3 min read
What decision does it make?
Each time someone looks at the locked phone, face unlock makes one yes-or-no choice: is this the owner? It does not try to work out who a stranger is. It only checks one face against one saved face.
| Decision | Possible answers |
|---|---|
| Is this the phone's owner? | Yes: unlock the phone. No: stay locked. |
What does it look at?
- The saved face. The phone keeps a saved picture of the owner's face to compare against later.
- The new picture. Each time the phone is picked up, the camera takes a fresh picture of whoever is in front of it.
- Facial features. The program compares the features of the two faces, not the background or the clothes.
- The match score. All the comparisons are added up into one number: how alike the two faces are.
How does it learn the rule?
- Learn from examples. Before it reaches a phone, the program learns from many example pictures of faces.
- Learn to compare. It learns to give two pictures a score for how much they show the same face.
- Save the owner's face. The phone saves a picture of the owner's face, so each new picture has something to be compared with.
- Set the line. People choose how high the match score must be. A high line keeps other people out, even if the owner sometimes has to try again.
Where can it go wrong?
- Someone else gets in. A person who looks a lot like the owner, such as a close relative, could get a match score over the line. This is a wrong yes.
- The owner is kept out. A dark room, a tilted phone or a big change in looks can push the owner's score under the line. This is a wrong no.
- It is not equally good for everyone. A government test lab found that many face matching programs make more mistakes for some groups of people than for others.
- No line is perfect. Move the line up and fewer other people get in, but the owner is kept out more often. Move it down and the reverse happens.
Try it
Toy example with made-up scores: move the line to decide how close a match must be to unlock. Can you keep everyone else out without locking out the owner?
A toy example: ten unlock tries with a made-up match score from 0 to 100, and whether the person really is the owner.
| Item | Values | Right bin |
|---|---|---|
| Owner, facing the camera in daylight | Match score: 95 | Unlocks |
| Owner, wearing new glasses | Match score: 85 | Unlocks |
| Owner, just woke up | Match score: 75 | Unlocks |
| Owner, phone held at a tilt | Match score: 65 | Unlocks |
| Close relative who looks alike | Match score: 60 | Stays locked |
| Owner, in a dim room | Match score: 55 | Unlocks |
| Friend with a similar haircut | Match score: 40 | Stays locked |
| Classmate | Match score: 25 | Stays locked |
| Stranger at the bus stop | Match score: 15 | Stays locked |
| Pet cat walking past | Match score: 5 | Stays locked |
Check it
What does face unlock decide?
Answer: Whether this is the phone's owner. It makes one choice: owner or not owner.
You move the line higher. What happens?
Answer: The owner is kept out more often. A higher line keeps others out better, but some of the owner's tries fall under it too.
A close relative unlocks the phone. What kind of mistake is that?
Answer: A wrong yes. The phone said yes, this is the owner, when it was not.
For older readers
This is called one-to-one face verification: one new image is compared with one enrolled image, and a similarity score is checked against a threshold. Letting in someone else is a false match; locking out the real person is a false non-match. Test labs measure the false non-match rate at a threshold chosen to give a set false match rate, because for security a false match usually matters more. Error rates can differ between demographic groups, which is a form of bias.
Questions
Your match score fell under the line, for example in a dark room or with the phone at an odd angle. Try again facing the camera.
It is possible. If their face scores over the line, the phone treats them as you. That is why the line is set high.
No. It only checks one question: does this face match the one face saved on the phone?
Not always. Tests by a government lab found that many face matching programs make more mistakes for some groups of people than others.
Sources
- NIST: Face recognition technology evaluation, 1:1 verification (opens another website)
- NIST: Testimony on facial recognition technology (verification, accuracy and demographic effects) (opens another website)
- University of Washington CSE 416: Classification (confusion matrix and changing the threshold) (opens another website)