Page 56 - ITU-T Focus Group Digital Financial Services – Technology, innovation and competition

ITU-T Focus Group Digital Financial Services
Technology, Innovation and Competition

3.2.3.3 Accuracy

A biometric system needs to have an acceptable level of accuracy that can be defined in an unambiguous
manner. As highlighted above, a biometric template is only a statistical representation of a physical feature,
not the feature itself; so a fingerprint template is not a fingerprint. By its nature, a statistical system is not
100% accurate and deviations from the ideal occur.
The accuracy of the biometric solution is measured by three metrics: (1) failure to enrol (FTE) rate; (2) false
rejection rate (FRR); and (3) false acceptance rate (FAR). These are described below. Although the accuracy
issues described may represent a small proportion, when scaled to a population, accuracy issues can affect a
great many people.

FTE

The FTE rate is the percentage of people who fail to be enrolled successfully into a biometric system.

A notable FTE problem is associated with fingerprint biometrics, because a percentage of the population have
unusable fingerprints for measurement due to imperfections, wear, or being an amputee. This is particularly
an issue with a population in a dry, dusty environment, manual labourers (including farmers), smokers, and
older members of society.

FRR

The FRR is defined as the percentage of verifications in which an incorrect verification or false rejection occurs
– that is, people whose attempt to verify themselves fails even though they are in fact the registered person.
For example, if the FRR is 0.1 per cent, it means that on average, out of every 1000 persons attempting to
access the system, one will not be recognised by that system.

It is important to note that the occurrence of an instance of false rejection may result in denial of service to
a valid user.

FAR
The FAR is defined as the percentage of verifications in which an incorrect or false acceptance occurs. For
example, if the FAR is 0.01 per cent, it means that on the average, one out of every 10,000 impostors attempting
to breach the system will be successful.
It is important to note that the occurrence of an instance of false acceptance may result in access to a service
being granted to the wrong person, in impersonation of another person.
Service requirements for accuracy

For any biometric technique, there is a direct relationship between the failure rates for FRR and FAR – that
is, decreasing one rate increases the other. This means that for a specific service a trade-off must be made
between the settings for FRR and FAR. The trade-off made will depend on the application; for example, in
financial services, it may be most important for rejections to be low, while in a government identity scheme,
the opposite may be true.
This trade-off is a decision that the organisation deploying the service elements must take. If a service is relying
on someone else’s biometric readers (for example, fingerprint readers built into mobile phone handsets) then
the service may not be able to decide on the balance between failure rates and may have to adjust their policies
to accept the reader owner’s selection.

42