Innovatrics’ latest latent fingerprint algorithm leads four accuracy measures in the National Institute of Standards and Technology’s Evaluation of Latent Friction Ridge Technology (ELFT). The innovatrics+0014 submission placed first for two hit-rate measures and two false negative identification rate measures in NIST’s current results.
ELFT is a one-to-many biometric evaluation of the algorithms used to create templates and search latent friction ridge images against large reference databases. NIST’s current table includes 22 active algorithm submissions from 12 vendors, with latent images or examiner-marked feature sets used as search probes and fingerprint records used as references.
Innovatrics recorded a 99.3 percent rank-1 hit rate without a similarity-score threshold, meaning the correct subject appeared first in the candidate list in 99.3 percent of searches. It also achieved a 98.9 percent rank-100 hit rate at a false positive identification rate (FPIR) of 10 percent, measuring how often the correct subject remained within the first 100 candidates after weaker results were filtered by a threshold.
On NIST’s DoD-Provided Dataset #1, the algorithm produced a 4.85 percent false negative identification rate (FNIR) at an FPIR of 0.01, equivalent to a one percent false positive identification rate. Its FNIR at rank 1 was 3.01 percent. Lower FNIR values indicate fewer searches in which the correct subject is missed under the specified threshold or rank.
The Defense Department dataset contains 5,257 operationally collected latent probes and an enrollment set of approximately 1.6 million non-mated subjects plus the corresponding mated identities. NIST provided no examiner-marked extended feature set data for those searches, so the algorithm processed the probe images directly.
The new results improve on the company’s preceding ELFT submission, which achieved a 98.6 percent rank-1 hit rate and a 98.2 percent rank-100 hit rate. On the same Defense Department dataset, FNIR at an FPIR of 0.01 fell from 5.19 percent to 4.85 percent, while FNIR at rank 1 declined from 3.23 percent to 3.01 percent.
Rank-based searches return a candidate list for review by a trained latent print examiner, while threshold-based searches can suppress weaker candidates before review. Innovatrics has also expanded the latent-print workflow in its criminal ABIS, adding multi-user examination tools and automated searches of new enrollments against unresolved evidence.
NIST evaluates the template creation and searching algorithms at the core of automated fingerprint identification systems, rather than complete deployed systems. The agency states that its publication of commercial test results does not constitute an endorsement of a company, system or product by the U.S. government.
Sources: Innovatrics, NIST ELFT Results, NIST Technical Report
–
By the ID Tech Editorial Team
