Biometric systems either confirm who you claim to be or search for who you are
A fingerprint scanner can work in two very different ways. In verification mode it compares your print with one stored template to confirm you are who you claim to be; in identification mode it searches a whole database to work out who an unknown person is.
Biometrics are measurements of the body and behaviour used to recognise people. Physiological traits include fingerprints, palm veins, faces, DNA, irises, retinas, hand geometry, scent and ear shape; behavioural ones, sometimes called behaviometrics, include typing rhythm, gait, signature and mouse movement. Because such traits belong to one individual, they confirm identity more reliably than a passport or password, though gathering them raises privacy concerns.
In 1999 Jain and colleagues proposed seven tests for judging any trait. Everyone using the system should have it; it should differ enough between people; it should stay stable over time; it should be easy to capture; the technology should be fast, accurate and robust; users should accept it; and it should be hard to fake. No single trait wins on every count, so the right choice depends on how much convenience and security a given application demands.
A system first enrols a person, capturing a sample through a sensor, cleaning it up, pulling out key features and saving them as a compact template while discarding the rest, which shrinks the file and helps protect the enrollee. Later samples go to a matcher that measures how far they sit from stored templates, for instance by Hamming distance, against a preset threshold. Identification also allows negative recognition, checking whether someone is a person they deny being, a job passwords or keys cannot do.
Single traits fail in predictable ways: irises age, and fingerprints wear down or get cut. Multimodal systems combine several, fusing them at the level of raw features, match scores or final decisions, with feature-level fusion expected to work best because it keeps the richest information. They are widely assumed to shrug off spoof attacks with fake traits more easily, but recent studies found that faking even one trait can defeat them.
Source: Biometrics
