Facial recognition: how it works, explained simply

In plain words

When your phone opens just because you looked at it, it is not looking at a photo of you the way a friend would. It measures your face, the shape around the eyes, nose, mouth and jaw, and turns that into a list of numbers often called a faceprint.

Then it compares. Your new faceprint is checked against the one saved when you set up the phone. If they are close enough, above a set threshold, it is a match. Nothing has to be identical, which is why it still works with a new haircut or in different light.

There are two kinds of use. Verification asks "is this the person they claim to be?", as with opening your phone or airport boarding. Identification asks "who is this?" by searching a large database, as some police systems do. The second raises bigger questions about privacy and mistakes, which is why several countries limit it.

Everyday example

The building security guard

A society's security guard does not memorise photos. Over time they notice what makes each resident's face their own: the eyebrows, the shape of the chin. When someone walks in, they compare quickly with what they remember and nod them through, or stop them if no one fits. Facial recognition does the same comparison, with numbers instead of memory.

Where the example breaks: A guard can ask a question or recognise someone from their walk. A face system only compares numbers. Twins, a poor camera angle, or a group that was thin in its training data can push it to a wrong match or a wrong refusal.

Go deeper

How it actually works

Modern facial recognition is a deep learning pipeline with four steps.

Detect the face
A detector, a form of image recognition, finds where faces are in the image and crops them.
Align it
Landmarks such as eye corners and the tip of the nose are located, and the face is rotated and scaled to a standard position.
Make a faceprint
A convolutional neural network turns the face into an embedding: a list of numbers. It is trained so that photos of the same person land close together and different people land far apart.
Compare against a threshold
For verification (one to one), the distance to one saved faceprint is checked. For identification (one to many), the closest matches in a database are returned. The threshold trades off false matches against false rejections.

Face ID odds: Apple says the chance a random person opens your phone with Face ID is less than 1 in 1,000,000, higher for twins, look-alike siblings and children under 13
A turning point: 2014: Facebook's DeepFace reached 97.35% on a standard face-matching test, close to human performance
Bias found: A 2019 NIST study of nearly 200 algorithms found false matches 10 to 100 times more often for some groups, depending on the algorithm
Rules: Since February 2025, the EU AI Act bans untargeted scraping of face images to build recognition databases and tightly limits live police use in public

To protect yourself: use opt-in services that keep data on your device, ask about alternatives when face scanning is offered, and check your phone's photo app settings for face grouping.

Should police use live facial recognition in public places?

One view

Use it, with safeguards

Supporters say it can find missing people and suspects faster, and that rules, audits and human review can keep errors and misuse in check.

The other view

Ban or strictly limit it

Critics say scanning everyone in a crowd treats all as suspects, chills free movement, and that error rates differ across groups, so the harm falls unevenly.

Most likely: Most rule-makers have landed in between. The EU bans it by default and allows narrow exceptions, such as searching for a kidnapping victim, usually with prior approval from a judge or independent authority.

Questions people ask

How does facial recognition work?

It finds the face in an image, lines it up, turns it into a list of numbers called a faceprint, and measures how close that is to saved faceprints. If the distance is below a threshold, it reports a match.

Is Face ID facial recognition?

Yes, a verification kind. Apple's Face ID projects thousands of invisible dots to map the face in 3D, compares it with your saved data, and keeps that data encrypted on the phone rather than in the cloud.

How accurate is facial recognition?

The best systems are very accurate on clear, front-facing photos. Accuracy drops with poor light, angles, masks, age changes and low-quality cameras, and NIST testing found that many algorithms err more for some demographic groups than others.

Is facial recognition used at Indian airports?

Yes. DigiYatra, launched at Delhi, Bengaluru and Varanasi airports in December 2022, lets passengers who register on its app pass checkpoints using their face. It is opt-in. The government says the data is stored on the passenger's own device, not in a central database. Critics have raised concerns about how the data is stored and shared, and the civil aviation ministry has said the not-for-profit body that runs it is outside the RTI Act.

What is the difference between face detection and facial recognition?

Face detection only finds that there is a face and where, like a camera's focus box. Facial recognition goes further and works out whose face it is, by comparing it with known faces.

Facial recognition