TECH5 Technology Blog

Introduction

Friction Ridge Image and Features (FRIF) Technology Evaluation (TE) Exemplar One-to-Many (E1N), or FRIF TE E1N, is a new open-set identification evaluation of algorithms launched in 2025 that automatically extract and use features from all types of exemplar friction ridge images (e.g., rolled fingerprints, palm prints, slaps) and later use those features to search for similar candidates in databases of millions of subjects. Formerly, it was known as FpVTE and was conducted previously in 2012.

Why NIST FRIF TE E1N was needed

FRIF E1N is a relaunch of previous evaluations conducted by NIST under the FpVTE moniker. This evaluation was highly needed, because the previous similar 1:N fingerprint NIST testing’ results were 13 years old and lost relevance.

The end customers are looking at these tests as a reference when selecting large-scale ABIS (Automated Biometric Identification System) platforms for their projects in civil identity, foundational identity, elections, passport systems and the like, and now can use the latest evaluation instead of the results from 2012.

How this evaluation works

NIST’s FRIFTE E1N gives an apples-to-apples, third-party audit of fingerprint feature extraction and search under tightly controlled conditions. NIST enforces a fixed “Timing Sample” with maximum processing times and runs all timed functions single-threaded on the same Xeon 6254 hardware, which makes speed claims comparable across vendors and closer to operational reality.

The track also exercises truly large enrollments (≈100k to 5M subjects) derived from the FpVTE 2012 datasets and reports DET/CMC accuracy, template sizes, and proprietary database footprints — so evaluators see not just accuracy but also storage and throughput implications at scale.

Success depends on end-to-end behavior across many capture types (index pairs, slaps, 10-print plain and rolled), sometimes with vendor-performed slap segmentation, correct finger/region “location” reporting, and strict speed caps while searching millions of subjects with a candidate list fixed at 100.

Building enrollment databases, meeting average search-time limits, and balancing FPIR vs. FNIR across these scenarios all must work simultaneously, and NIST publishes only submissions that meet the speed requirements. Performing well therefore demonstrates not just an accurate matcher but a production-ready pipeline: robust feature extraction and segmentation, efficient indexing/search, sensible thresholds at low FPIR, and optimal storage at national scale.

1-N fingerprint matching powers “who is this?” lookups at scale: national IDs and voter rolls (deduping enrollments), border/visa screening (watchlists), and law-enforcement AFIS (arrestee/latent ID). It also runs big private systems such as badge-less access and timekeeping, telecom and bank e-KYC, healthcare patient matching, and mobile kits for field registration in emergencies or benefits programs.

Turning it into production means more than a fast matcher: you need reliable capture and quality control, strict latency/throughput at peak, compact templates and scalable databases, and thresholds tuned to risk (e.g., ultra-low FPIR at borders vs. higher recall with human review for dedupe). Add fallbacks for poor fingers, auditability, privacy/retention controls, and monitoring for drift. Strong FRIF results signal you can meet those demands in the wild — accurate, fast, and scalable at national or enterprise scale.

Results

Because this evaluation is highly complex, and since NIST publishes only submissions that meet the required speed benchmarks, only two biometric vendors have passed the test so far. Although NIST does not currently provide a consolidated results table, individual performance metrics are publicly available on the official NIST website.

See results

TECH5 submitted its algorithm to NIST FRIF TE E1N on August 11th, 2025. In head-to-head testing, TECH5 outperforms competition on both accuracy and efficiency across two out of three datasets: Class A (index fingers) and Class C (ten fingers). On Class A’s 1.6M both-index search, TECH5 posts a lower miss rate at an operational threshold (FNIR @ FPIR≤0.001: 0.001 vs. 0.003) and a better Rank-1 FNIR (0.0004 vs. 0.0005), showing higher accuracy at the top of the candidate list. On Class C’s 5M ten-finger searches, TECH5 shows better results across all pairings: Plain-Plain (0.0018 vs. 0.0039), Plain-Rolled (0.0006 vs. 0.0024), and Rolled-Rolled (0.0003 vs. 0.0031), and leads on Rank-1 FNIR (0.0042/0.0037/0.0034 vs. 0.0061/0.0050/0.0055), demonstrating stronger identification performance at scale.

TECH5’s algorithm has a higher template creation speed: ~2× quicker (e.g., Identification Flats mean ~3.7–3.8s vs. 9.0–9.1s) and templates are smaller (e.g., Ten Fingers Plain ~29 kB vs. ~40 kB), which translates into lower compute and storage costs. At deployment-scale database sizes, TECH5’s enrollment footprint is smaller: e.g., 5M Ten Fingers (Plain) ~148.9 GB vs. 173.6 GB; 5M Ten Fingers (Roll) ~305.6 GB vs. 339.1 GB; 3M Identification Flats ~88.8 GB vs. 98.4 GB — supporting more economical, high-throughput operations.

TECH5’s technology, used for this evaluation, is part of T5-OmniMatch ABIS multi-modal biometric matching platform, providing access to state-of-the-art performance to company’s partners and customers.

Learn more

Subscribe to the TECH5 Blog