Biometric template

The mathematical summary of a face, fingerprint, or other biometric trait, stored instead of the original image or recording.

Instead of keeping a person's photo, a system keeps a string of numbers that captures the features it uses to recognize them.

A feature vector derived from a biometric sample and used for comparison, designed to be hard to turn back into the original image. A feature vector is an ordered list of numbers that describes the distinguishing features of a face or other trait.

A biometric template is created and used in a few steps:

1. The system captures a sample, such as a selfie.
2. An algorithm extracts the distinguishing features and converts them into a template.
3. The system stores the template as the reference at enrollment.
4. Later captures are converted into templates and compared with the reference.

Templates have a few notable properties:

• Algorithm-specific, meaning templates made by one algorithm usually cannot be compared with templates made by another.
• Compact, which makes comparison fast.
• Protectable, meaning systems can encrypt templates and limit who can access them.

Biometric templates sit next to terms that are easy to mix up:

• Biometric sample is the raw capture, such as the photo, while the template is the numeric summary.
• Face match compares two templates and produces a similarity score.
• Data minimization is the privacy principle that favors keeping only what is needed, which is one reason to store a template instead of an image.

For more, see Biometrics and the Looming Quantum Threat and the Privacy Hub.