More Than 25 Years of Independent Biometric Performance Testing

First place. VeriFinger achieved the best reliability results among all participants.

• The first Fingerprint Verification Competition, entered under the company's former name, Neurotechnologija.
• Marked the beginning of participation in independently organized fingerprint algorithm competitions.

One silver and two bronze medals across the evaluated datasets and measures.

• Continued participation with a fingerprint recognition algorithm across standardized FVC datasets.

One of the strongest reliability results in the Middle Scale Test.

• First NIST fingerprint evaluation, entered as Neurotechnologija.
• FpVTE assessed one-to-many identification and gave an early independent measure of suitability for larger databases.

VeriFinger took four gold in the Open Category; FingerCell six silver in the Light Category.

• VeriFinger: four gold, three silver and two bronze medals in the Open Category.
• FingerCell, for embedded and lower-resource platforms: one gold, six silver and three bronze in the Light Category.
• The two categories demonstrated fingerprint algorithms for both full-scale systems and constrained platforms.

First place. The P058 algorithm ranked highest overall in the Open Category by Average Zero FMR.

• Four gold, two silver and two bronze medals in the Open Category.
• A separate algorithm ranked second in the Light Category by Average Zero FMR, taking one gold and four bronze medals.
• Zero FMR measures performance where false acceptance is reduced to zero – the operating point that matters for high-security deployments.

Recognized MINEX compliant for both template generation and matching.

• MINEX tests the interoperability of fingerprint templates created and matched by different implementations.
• Relevant wherever biometric records must stay usable across equipment, software versions or organizations.

Seven FBI certifications, one for each supported computing platform.

Placed second in the Ongoing MINEX ranking.

Ranked fourth for overall accuracy; four times more accurate than the second-fastest algorithm in Class C.

• MegaMatcher SDK and MegaMatcher Accelerator were evaluated on single fingers, flat tenprints and complex tenprint records.
• The Class C result was measured on 5 million tenprints under a 90-second search limit.

Entered the ongoing MINEX III template interoperability program.

• MINEX III remains active, so results may change as participants submit updated algorithms.

MegaMatcher On Card passed MINEX III testing.

• MINEX III is the ongoing evaluation of fingerprint template interoperability for the US PIV program.
• MINEX compliance is often a mandatory requirement in public tenders in the United States and elsewhere.

First place. Most interoperable matcher, and fourth most accurate native template matcher.

Submission 4E ranked among the most accurate algorithms across all 33 experiments.

• Second fastest during enrollment and second or third fastest in template matching.
• PFT II evaluated one-to-one verification with proprietary templates on datasets of up to 120,000 subjects; the program ended in 2019.

First place. First for template generator interoperability, second for matching.

First place. Took the top position on entry to the program.

• PFT III measures one-to-one fingerprint verification using proprietary templates, with successive algorithm versions submitted over time.

First place. First place in the template generator category.

Active evaluation. Evaluated for separating individual fingerprints from four-finger slap captures.

• Published results show tradeoffs between speed and accuracy; the submission was neither the fastest nor the most accurate across the four reported categories.
• Slap segmentation matters in enrollment systems that capture four-finger slaps – national ID, voter registration, border control and criminal registration workflows.

First place, active evaluation. Top result in the standard fingerprint verification benchmark.

• FVC-onGoing accepts new algorithms continuously while keeping the evaluation datasets fixed, so results stay comparable over time.

First place, active evaluation. Top result in the hard-conditions verification benchmark.

• FV-HARD uses the same protocol as FV-STD but on more difficult image conditions.

First place. Top position confirmed in the 2022, 2023, 2024 and 2025 releases.

Smart-card submission 020A confirmed MINEX III compliant.

First place. VeriFinger 13.1 ranked first among the valid submissions.

• Highest area under the ROC curve at 0.854, with an equal error rate of 0.228.
• Failure-to-enroll fell to 6.2% from 40.8% for the VeriFinger 12.3 baseline – not the best FTE in the competition.
• Lowest computational cost of all submissions: 0.78 seconds per fingerprint pair.

First place. Highest accuracy across most of the evaluated datasets.

• December 2024 results: highest accuracy on most evaluated datasets and among the leaders for feature extraction speed.
• ELFT measures latent friction ridge identification – marks recovered from crime scenes searched against ten-print and palm databases.

First place. First place for fingerprint recognition out of 2,106 applications.

• Strongest one-to-one matching performance among the three submissions selected for final evaluation.
• Used a longitudinal dataset: samples from children aged 5–10, recaptured 5–10 years later, testing age invariance.

First place, active evaluation. First in the majority of test categories, including index-finger and multi-finger identification.

• Successor to FpVTE 2012 for large-scale one-to-many fingerprint identification.
• Neurotechnology+0106 tied for first in Class A index-finger identification and achieved zero errors in four experiments.
• Class B: second of 11 on individual slaps. Class C: second in ten-finger identification.
• Against 5 million records, median search times were 613 ms plain and 560 ms rolled, with zero API timeouts.

Participated in both the recognition and the new quality-assessment tracks.

• The second edition split the evaluation into latent fingerprint recognition and latent fingerprint quality assessment.
• Neurotechnology co-authored the publication presenting the competition and its results.
• [INTERNAL CONFIRMATION REQUIRED: Add Neurotechnology's exact 2025 recognition and quality-assessment placements after checking the complete IJCB paper. Until then, state participation only; the results are published, but the company's placement has not been confirmed from the paywalled paper.]

Fastest performance and near-best accuracy in most categories.

• A top performer in the 2025 SlapSeg III results across most evaluated categories.
• [INTERNAL CONFIRMATION REQUIRED: This statement conflicts with the SlapSeg III entry above, which reports that the submission was neither the fastest nor the most accurate across the four reported categories. Confirm which result release each statement describes before publication.]

First place, active evaluation. First place in both identification and investigation scenarios.

• First in both scenarios on the IARPA N2N and FBI-Provided Solved #1 datasets.
• Leading positions on the DoD-Provided Dataset #1 and shared first place on the FBI Laboratory dataset.

First place, active evaluation. Most interoperable PIV Level 1 generator and most accurate native Level 2 generator and matcher pair.

• Also the leading interoperable Level 1 generator and matcher pair.

First place, active evaluation. Highest accuracy, alongside a second submission tuned for speed and template size.

• One submission achieved the highest accuracy; another emphasized faster matching and smaller templates.

First place. First place in the palmprint recognition benchmark.

Among the top eight most accurate algorithms out of 16 vendors.

• An experimental face engine integrated with a face tracker, tested on passively recorded non-cooperating subjects.
• Findings later contributed to improvements in VeriLook SDK and MegaMatcher SDK.

Sixth of 40 participants at 12 million subjects; third of 40 on images captured up to 18 years after enrollment.

Evaluated in biometric boarding and airport security gallery simulations.

• Tested one-to-many identification against galleries built from passport, visa or previous travel images.

Among the leading algorithms in border-control, kiosk, frontal and profile mugshot scenarios.

• Published 13 January 2022, covering both one-to-one verification and one-to-many identification.
• In July 2023 FRVT was split into FRTE and FATE, and participation continued in both tracks.

Active evaluation. Leading accuracy in supervised and unsupervised visa, border and kiosk scenarios.

• Published 27 November 2023.
• FRTE 1:1 measures one-to-one face verification across a range of operational image sources.

Active evaluation. Strong results in civil identification and investigation scenarios.

• Includes both frontal and profile mugshot matching.

Active evaluation. Top results for recognizing masked and otherwise occluded faces.

• Published 20 December 2023.
• Measures the accuracy cost of lower-face occlusion, which became an operational requirement for many deployments.

Active evaluation. Facial image quality assessment tested on challenging photographic conditions.

• Quality estimation matters because enrollment and matching accuracy depend on pose, focus, lighting, resolution and facial visibility.

Active evaluation. Evaluated for detection of face morphing attacks.

• Tests detection of images combining the features of two or more people, which could otherwise be used by more than one person.
• Relevant to document issuance and identity fraud prevention.

Face presentation-attack detection certified at Level 2.

• An independent certification, separate from NIST FATE PAD – it should not be presented as participation in that NIST evaluation.

First place. DMSTI-Neurotechnology ranked first in the Full Data Track.

• A research collaboration with DMSTI, the Institute of Data Science and Digital Technologies at Vilnius University.
• Adapted CLIP ViT-L/14 on approximately 3 million synthetic face images across 35,000 identities, using full fine-tuning with a Sub-Center ArcFace head.
• Highest average accuracy on small-scale benchmarks at 95.51%, leading IJB-B and IJB-C at all reported FAR thresholds.
• 87.42% TAR at FAR 10⁻⁵ on IJB-C versus 76.71% for the FRoundation baseline; led RFW at 91.70% average accuracy with the lowest standard deviation, 2.33.

VeriEye 2.1 ranked among the top three for accuracy, 25–75× faster than competitors.

• Templates were 3.5–7.5× smaller than those of competing algorithms.
• Established the starting point for independently evaluated iris recognition development.

Second-fastest matcher, with three times the accuracy of the only faster contender.

• NIST evaluated more than 86 algorithms and variations from nine companies and two universities.
• IREX III focused on performance and interoperability of compact iris templates.

The only submission returning results in under 0.5 s on a 1.6-million-iris gallery.

• 66 iris prototypes from 12 commercial and academic organizations were evaluated in a large-scale identification task.
• Across all tests, submissions ranked among the top four participants for accuracy.

Ranked second for accuracy; the accelerated version was nearly 50× faster than any other matcher.

• NIST evaluated 46 iris recognition algorithms from 13 commercial and research organizations.

Joined the ongoing large-scale iris identification evaluation.

• IREX 10 accepts submissions continuously and measures one-to-many iris identification at scale.

First place. Judged the most accurate participant in the Rank 1 category.

• Outperformed other contenders in both single-eye and two-eye assessments.
• Top results across most of the reported performance metrics.

First place, active evaluation. Ranked first in all IREX 10 categories, single-eye and two-eye alike.

• Lowest Rank 1 miss rate in both two-eye and single-eye testing.
• Lowest false negative identification rate at a false positive identification rate of 0.01.
• Top results for both automated identification and Rank 1 investigation-style accuracy.

Participated in an evaluation series NIST has run since 1996.

• SRE24 assessed automated person detection using conversational telephone speech and audio from video.
• Audio-only, visual-only and audio-visual tracks, with cross-source and cross-language comparisons.
• NIST names participating teams but does not link them to scores, and restricts promotional claims about standing.