Youverse highlights population-scale facial recognition accuracy in NIST evaluation Youverse says its first submission to the NIST’s ongoing Face Recognition Technology Evaluation 1:N Identification track validates the accuracy of its software as suitable for population-scale fraud detection. The company’s youverse_001 facial recognition algorithm recorded a false-negative identification rate of 0.43 percent at a threshold limiting the False Positive Identification Rate (FPIR) to approximately 0.3 percent when matching frontal mugshot probes against a gallery of 1.6 million mugshots. Youverse expresses this as a 99.57 percent true identification rate. “Independent evaluation by NIST is a major milestone for Youverse and a powerful validation of what we have built,” says Miguel Lourenço, the company’s CPO. “Achieving a 99.57 percent true positive identification rate against a gallery of 1.6 million faces demonstrates that our technology delivers the scale and accuracy demanded by the world’s most critical identity systems,” Lourenço adds. One-to-many identification compares one probe image against an entire gallery to find a possible matching identity. Organizations often use 1:N identification to search large databases for duplicate enrollments or people registered under different identities. When searching a NIST mugshot gallery with 12 million identities, the algorithm recorded a 98.97 percent identification rate. Youverse says this large-scale search capability could support identity deduplication, fraudulent enrollment detection, access control, KYC, and anti-money laundering (AML) processes. The company’s best result by ranking among the main FRTE metrics was matching probe images captured by biometric kiosks against a gallery of 1.6 million visa images. Youverse delivers the facial matching algorithm with liveness detection and says its platform permits one-to-many searches only with explicit user consent. The latest NIST FRTE 1:N results showed that performance continues to converge in controlled comparisons such as frontal mugshot identification. Important performance differences remain when algorithms face more difficult images, including webcam captures,