Neurodactyl releases high-speed biometric deduplication tool for large-scale databases
Neurodactyl, a Georgia-based tech startup specializing in contactless fingerprint recognition, has unveiled a new biometric deduplication tool that quickly identifies errors and duplicates in large-scale biometrics databases.
A distinctive feature is its direct all-to-all matching, where every fingerprint is compared to every other in the database, regardless of finger position.
In practical terms, for a database containing ten million people and 100 million fingerprints, this translates to a staggering five quadrillion matching operations. Neurodactyl’s deduplication algorithm uses neural networks to create compact biometric templates that can be quickly matched, according to the company announcement. Through batch processing, the deduplication software can perform up to a billion matching operations per second on a single CPU, and times as many on a GPU.
That means the tool can deduplicate a database with 100 million fingerprint templates in less than ten days, using a single computer with a high-end consumer GPU. If relying solely on a CPU, the process takes approximately two months.
Efficient deduplication of biometric databases helps to streamline authentication processes and enhances biometric database security and reliability. It is considered important for preventing fraud in applications from public benefits payments to voting.
Neurodactyl also passed the NIST PFT (Proprietary Fingerprint Template) III evaluation earlier this year.