False rejection rate (FRR) is a security metric used to assess the performance of biometric systems such as fingerprint recognition, face recognition, or iris scanning. It refers to the frequency that a system mistakenly rejects an authorized user as an impostor. In statistical terms, FRR is a Type I error. It is important to note that while a low FRR is desirable, it must be balanced against a high False Acceptance Rate (FAR) which measures the number of times the system mistakenly accepts an illegitimate user as a valid one. Often the FAR and FRR will be equal, and in this case the system is considered to have excellent performance
The higher the FRR of a system, the more secure it is. This is because a true impostor will fail to match the template of an authenticated user and will be rejected. However, a low FRR can also lead to poor usability of the system, as legitimate users will be required to enter their credentials several times in order for the system to correctly recognize them.
In practice, FRR can be optimized by adjusting the decision threshold of a system. A system with a higher decision threshold will have lower FRR, while a system with a lower threshold will have higher FAR. A good threshold is one that can be reliably used by a large majority of the population while still being accurate enough to detect a few false positives. These parameters can be graphically represented using an ROC curve which is widely used in psychophysical evaluations of sensory systems.