False rejection rateFRR

How often the system wrongly rejects the right person.

If a face check has a false rejection rate of two percent, about two out of every 100 genuine customers would be turned away even though they are who they say they are.

The proportion of genuine comparisons incorrectly rejected at a given threshold. A genuine comparison pairs a sample with a record of the same person, and the threshold is the similarity score below which the system declares no match.

The rate depends on the threshold:

• A stricter threshold rejects more borderline cases, so the false rejection rate rises.
• A looser threshold accepts more borderline cases, so the rate falls.

Several things push genuine comparisons below the threshold:

• Poor image quality, such as blur or low resolution.
• Lighting and pose, such as glare or an angled face.
• Changes over time, such as aging, a new hairstyle, or glasses.
• Demographic differences, since error rates can vary across groups.

NIST and many standards report the algorithm-level version as the false non-match rate (FNMR), and false rejection rate (FRR) is the familiar name for the same trade-off.

False rejection rate sits next to terms that are easy to mix up:

• False acceptance rate (FAR) is the opposite error, where a different person is wrongly accepted.
• Equal error rate (EER) is the single point where FAR and FRR are equal.
• Conversion rate is the share of people who finish a flow, which false rejections reduce.