Latest NIST Test Shows Cognitec’s Face Recognition Algorithm with Overall Best Accuracy-Speed Tradeoff


Algorithm development at Cognitec continues to engineer the optimal balance between speed and accuracy of face matching processes.

The latest results of the U.S. NIST Face Recognition Vendor Test for identification tasks show the Cognitec algorithm in the best position of all algorithms when relating the template generation time to the false negative identification rate (miss rate) for mugshot databases.

The identification test addresses the largest market for face recognition applications, including detection of duplicates in image databases, and fraud detection during passport and driver’s license applications.

These tests apply a very high matching threshold, where only 0.3 % (3/1000) of probes without a mate in the gallery produce a false hit—one of the most difficult face recognition tasks.

“We are proud to also see significant accuracy advances in comparison to the algorithm we submitted to the last test in early 2021,” says Dr. Thorsten Thies, Director of Algorithm Development.

“For the test with 12 Million mugshot images, Cognitec achieved rank 24 of 165 algorithms with a match rate of 98.5 %. In the test with 1.6 Million mugshots, with a 99.4 % match rate, we ranked 26 of 299 algorithms. These results show remarkable performance consistency, regardless of database size.”


Comments are closed.