The latest Innovatrics submission to the NIST FRTE 1:N benchmark has resulted in outstanding progress. By nearly halving our error rate, we have advanced to the 11th-13th position out of 152 vendors and are among the select few with algorithms for iris, fingerprints and facial recognition all among the global top.
The latest Innovatrics facial recognition algorithm features several enhancements designed to reduce the identification error rates. Among these were the improved detection model, a revised approach to final score computation and re-ranking algorithm. All of these enhancements together have allowed our newest facial identification algorithm to jump up to the top decile of the global leaderboards.
“The benchmark includes various categories and scenarios, but our focus is on two specific areas that are the most important to our clients,” explains Jakub Sochor, head of Innovatrics R&D team. The first benchmark, VISA-BORDER, involves comparing faces captured during border crossings with visa application photos in the gallery. The second area, MUGSHOT-MUGSHOT, has two galleries of 1.6 million and 12 million photos. In all of these scenarios our algorithm has achieved significant gains.
Our latest results demonstrate substantial improvements in both scenarios, as indicated in the table below (via NIST):
Results shown from NIST do not constitute an endorsement of any particular system, product, service, or company by the U.S. Government
(Error rates are FNIR@FPIR = 0.003 (lower is better) as reported by NIST)
The individual rank for every single scenario is indicated in the parentheses in the superscript. We plan to integrate all of these improvements into production in the coming months.
NIST FRTE 1:N benchmark is the continuation of the former FRVT tests. FRTE is concerned with the accuracy of the face recognition, while FATE branch benchmarks face analysis technology.
Our passion for technology brings biometric benefits to millions of people.
