Home Identity Glossary Face Recognition

Face Recognition

What is face recognition?

Face recognition is the automated use of facial biometric data to compare a detected face with one or more face references. It can support verification, which compares a person with one specified reference, or identification, which searches for a person in a gallery of enrolled faces.

The term is broad, and vendors sometimes use it for an entire biometric product that includes detection, quality assessment, comparison, and liveness checks. Technical requirements should therefore state whether the intended operation is one-to-one (1:1) verification or one-to-many (1:N) identification.

How does face recognition work?

A face-recognition system commonly performs these stages:

Capture. The system receives an image or video frame containing a face.
Detection and alignment. It locates the face and prepares a standardized crop.
Template creation. An algorithm converts facial features into a biometric template.
Comparison. The system compares the probe with reference data under the selected mode.
Result. A threshold and business policy produce a match decision or a ranked list of candidates, depending on the comparison mode.

The comparison result does not establish whether a live person supplied the probe. Liveness detection can form part of presentation-attack detection at the camera, while injection controls address untrusted media introduced into the software process.

What are the main types of face recognition?

1:1 face verification tests whether a probe matches the reference linked to a claimed identity.
1:N face identification searches an enrolled gallery and returns one or more possible matches.

A 1:N result may provide a candidate instead of a final identity decision. Gallery size, threshold, ranking rules, and human review affect how the result is used.

What is face recognition used for?

Organizations use 1:1 verification for customer onboarding, account recovery, access control, and high-risk transactions. They may use 1:N identification for duplicate-account detection, biometric watchlist searches, investigations, and identifying an enrolled user who has not supplied an account identifier.

Because the two modes carry different error and privacy risks, contracts and policies should name the mode, reference source, threshold, gallery, and review process.

How accurate is face recognition?

Accuracy depends on the algorithm, comparison mode, threshold, image and reference quality, gallery composition, and capture conditions. A single percentage cannot describe performance for every use.

A 1:1 test commonly reports false match rate (FMR) and false non-match rate (FNMR). Open-set 1:N identification commonly uses false positive identification rate (FPIR) and false negative identification rate (FNIR). Testing should reflect the intended users and operating conditions, including camera type, lighting, pose, image age, and gallery size.

How can Regula help with face recognition?

Regula Face SDK supports face detection, 1:1 face matching, 1:N face identification, face-image quality assessment, and active or passive liveness checks. A 1:1 comparison can use a portrait from an identity document or RFID chip, a previous selfie, or another trusted reference. The identification module can search a captured face against a managed database of enrolled identities.

For a document-based 1:1 comparison, Regula Document Reader SDK authenticates the source document, extracts personal data and portraits, and reads supported NFC/RFID chips.

What is the difference between face recognition and face detection?

Face detection finds and locates faces in an image. Face recognition uses the detected facial data in a biometric comparison with reference data.

Can face recognition accuracy differ between demographic groups?

Yes. Error rates can differ between demographic groups, and the differences vary by algorithm, image quality, comparison mode, and threshold. Organizations should examine subgroup performance with test data that reflects the intended user population.

Can face recognition identify anyone from a photo?

No. A 1:N system can search only its available gallery, so the person needs an enrolled reference there. Image quality, threshold settings, gallery size, and algorithm performance also affect the search. A returned candidate may require further evidence or human review.

Can a photo or deepfake fool face recognition?

Yes. A comparison engine may return a high similarity score for a printed photo, replayed video, or synthetic face that resembles the reference. Presentation-attack detection examines threats shown to the camera, while secure capture and injection controls address media introduced directly into the software process.

Does face recognition store people’s photos?

It may. Some systems retain source images, some retain biometric templates, and some retain both for a defined period. Organizations should document the purpose, legal basis, access controls, retention period, and deletion process for every retained item.