1. Background

As IT has permeated people’s everyday life and work to a greater and greater extent, identity verification has been becoming very important. In particular, following the COVID-19 pandemic, people have been facing significant changes in lifestyles and work styles and having more opportunities to use biometric verification (identity verification using the face, fingerprints, veins, and iris) to ensure safety and security in the workplace, overcome labor shortages, and increase added value, and this has been increasing the demand for more accurate biometric verification.

Looking at the conventional ISO standard for methodologies for evaluating the accuracy of biometric verification, companies have been making use of ISO/IEC19795-1 (Biometric performance testing and reporting). However, as companies attempt to increase the accuracy of biometric verification, they need a very large number of samples, and this bottleneck hinders their improvement of the accuracy of biometric verification. To overcome this, Japan filed a proposal with the ISO for the development of a new international standard for methodologies for evaluating the accuracy of biometric verification using a small number of samples.

In response, the ISO initiated and advanced discussions on this proposal, which was led by Japan, and ultimately issued a new international standard ISO/IEC5152 (Biometric performance estimation methodologies using statistical models) in July 2024.

2. Outline of the international standard

This standard aims to help companies to apply methodologies in which, when evaluating the accuracy of biometric verification, they make use of extreme value statistics¹ for estimating the probability of the occurrence of rare events to evaluate a false acceptance² rate and a false rejection³ rate, thereby allowing companies to reduce the number of samples to one-third or less in evaluating the accuracy.

According to the conventional international standard for methodologies for evaluating such accuracy (ISO/IEC19795-1), companies need 2,450 or more samples to evaluate a false match rate of 0.0001% (1 in 1 million) and 7,746 or more samples to evaluate a false match rate of 0.00001% (1 in 10 million). However, the new international standard allows companies to reduce the number of samples for such evaluation to one-third or less, and, accordingly, using this standard, companies can estimate the performance of biometric verification systems with a false match rate of 0.00001% (1 in 10 million) only by using 2,450 samples, which is the number of samples required for an evaluation of a false match rate of 0.0001% (1 in 1 million).

photo

3. Expected effects

The new international standard is expected to help companies, if they apply the standard, to shorten the cycle for improving the performance of biometric verification systems, thereby allowing these companies to make use of biometric verification systems with higher accuracy. The standard contributes to ensuring safety and security in the workplace, overcoming labor shortages, and increasing the added value of various services, involving work style reforms (e.g., telework), educational reforms (e.g., online classes), online medical care and monitoring, unmanned stores, and online payments.

Notes

1. The term "extreme value statistics" refers to a statistical method for estimating the probability of the occurrence of rare events. This method has also been used to estimate the probability of the occurrence of natural disasters (e.g., large tsunamis), and the application of this method to biometric verification has made it possible to reduce the number of evaluation samples.
2. The term "false acceptance" refers to a mistake in which a biometric verification system accepts a different person as the target person. A higher false acceptance rate may lead users to consider the result of the identification process unreliable and stop using the system.
3. The term "false rejection" refers to a mistake in which a biometric verification system rejects the target person by identifying the person as a different person. A higher false rejection rate may lead users to consider the system inconvenient and stop using it.

Division in Charge

International Electrotechnology Standardization Division, Innovation and Environment Policy Bureau