Алгоритм TECH5 установил рекорд скорости поиска в базе данных из 500 тысяч лиц на тестировании NIST

TECH5 posts fastest search time in NIST IREX 10 iris evaluation TECH5 has recorded the fastest search time in the current two-eye developer table of the National Institute of Standards and Technology’s Ongoing Evaluation of Iris Recognition (IREX 10). IREX 10 evaluates one-to-many iris recognition, which searches for one person in a database of enrolled identities. The TECH5 biometric algorithm searched a gallery of 500,000 enrolled identities in 0.088 seconds and generated a 6,237-byte enrollment template. It recorded an FNIR of 0.0067 at an FPIR of 0.01. “With this NIST IREX 10 submission, we improved the identification accuracy of our iris algorithm while recording the lowest search time in NIST’s two-eye developer table and a notably compact template,” says Rahul Parthe, co-founder, chairman and CTO at TECH5. In a press release, TECH5 compared its results with the five highest-ranked submissions for two-eye accuracy, saying those algorithms require between 4.6 and 12 seconds to search the same 500,000-person gallery. The company extrapolates the template sizes to a hypothetical database of 100 million people. TECH5 estimates that its raw templates would occupy approximately 624GB when compared with 1.83TB and 13.09TB for the five other algorithms. These figures exclude database indexes, backups, redundancy, and other infrastructure. The reported 0.088-second figure measures only the database search under NIST’s test conditions. Operational deployments must also capture iris images, assess quality, perform presentation attack detection (PAD), generate biometric templates and return candidates for automated or human review. An earlier TECH5 submission also recorded the highest matching speed among IREX 10 participants at the time. The latest submission continues that emphasis on search speed, although other entries provide lower error rates. The results illustrate the tradeoffs among search speed, matching accuracy and template size. While TECH5 leads on search speed, Neurotechnology currently tops the reported accuracy categories, and Idbio has posted a lower false non-identification rate with a slightly slower search time and smaller template. TECH5 says its latest iris recognition algorithm will be available through the T5-OmniMatch product family for border control and national identity applications. Article Topics accuracy | algorithms | biometrics | IREX | iris biometrics | NIST | TECH5