Rank One Shows Significant Accuracy Gains in Latest FRVT Test

Rank One Computing (ROC) is boasting about the improved accuracy of its latest facial recognition algorithm. In that regard, the company noted that the algorithm has already been vetted by an independent source, posting a dramatic 30-40 percent accuracy gain in the most recent iteration of the NIST’s FRVT Ongoing test.

In absolute terms, ROC’s algorithm had a 99.21 percent overall accuracy rate in situations in which the False Match Rate is set at 1 in 1,000,000. The algorithm also scored well in several other categories that measure performance and efficiency. Most notably, ROC was in the 95^th percentile for template size and comparison speed, which is to say that the algorithm was able to achieve a high level of accuracy while using smaller (and more efficient) face templates.

According to ROC, its biometric templates are 20 times smaller than those generated with algorithms of a comparable quality. Rank One was one of the more consistent facial recognition vendors, and reached the 80^th percentile in seven of the eight benchmarking metrics that the NIST uses to evaluate accuracy and template generation. ROC noted that the NIST tested the algorithms using a diverse dataset with images of people from all over the world.

While the company is celebrating its current performance, Rank One indicated that it will not rest on its laurels moving forward. The company has already announced that it will release a new version of its facial recognition SDK in early 2022. The new version of the algorithm is expected to be even more accurate than its predecessor.

Rank One’s algorithm has scored well in template generation and comparison speed in prior FRVT tests. The company patented its liveness detection technology in early 2021, and it has since teamed up with the SIA to launch an industry campaign to create an ethical code of conduct for facial recognition developers.

–

September 17, 2021 – by Eric Weiss