In an industry fiercely bent towards competition and innovation, we are proud to share that NEC’s facial recognition algorithm again placed first for accuracy in two key categories of the National Institute of Standards and Technology’s (NIST) most recent Face Recognition Vendor Test (FRVT) 1:N Identification.
- #1 in Mugshot Identification
- #1 in Border Security Identification
This builds on two decades of success in which NEC has consistently outperformed the competition in terms of both accuracy and speed, even when faced with demographic differentials like race and sex, or when subjects are wearing facemasks for coronavirus.
The new NIST results published August 5, 2021, showed NEC’s algorithm outperformed the entire field of 291 entries submitted from all over the world on both mugshot identification and border security identification. Our NEC-004 algorithm demonstrated False Negative Identification Rate (FNIR) of just 0.0022 and 0.0345 at a False Positive Rate (FPIR) of 0.003, respectively.
These are the two most important categories for many NEC customers who rely on our technology for law enforcement, immigration, border security, access control for passengers boarding aeroplanes, for service members at military installations, for employees at government facilities, as well as for visitors of public venues like theme parks and stadiums.
The results from this week were an update to the NIST Interagency report 8271 originally published in September 2019. The major finding of that report was that massive gains in face recognition accuracy had been achieved in the years 2013 to 2018, and these far exceed improvements made in the prior period, 2010 to 2013. Here again today we can see the progress continues for NEC and it reconfirms the ability of face recognition technology to improve the user experience and serve as an effective tool for enhancing security for a wide variety of applications.
Over the last decade, our algorithms have consistently placed at the top of NIST’s ranking on a number of factors primarily due to our massive investments in research and a sharp focus on the needs of our customers. Ensuring algorithms are accurate is essential to helping our customers fulfil their important missions and building public trust in the use of these technologies.
In particular, assessing algorithms’ accuracy across demographic groups is a key component of efforts to mitigate the risk of bias and discrimination. We appreciate NIST’s work in this area and are proud that, when NIST evaluated face recognition algorithms’ performance across demographic groups in 2019, it found that NEC’s algorithm showed “undetectable” false positive error rate differentials across demographic groups based on race and sex.
In the era of COVID-19, NEC also demonstrated the ability to verify identity on subjects even when wearing facemasks. In testing performed by the U.S. Department of Homeland Security (DHS) at its “S&T Rally” conducted last year in Maryland, the NEC algorithms again outperformed all competition in several key categories. In fact, we showed a True Identification Rate of 98.7 per cent in scenarios that simulated a boarding gate at an airport with passengers wearing masks.
Throughout many years of NIST testing, the U.S. government has regularly benchmarked the entire facial recognition algorithm market against NEC’s industry-leading results. In 2019, NIST found that 19 companies had caught up to the NEC 2013 algorithm in overall accuracy. In 2020, a handful of companies were able to produce algorithms that met or exceeded NEC’s past performance. Here again, NEC is setting a higher bar.
We are especially proud to bring home first place honours in these evaluations because it affirms our mission to serve as a trusted partner to our country’s public servants.
To read more, you can access the full NIST report.
This article was first published on NEC Today on 9 August 2021.