biometrics.live builds face, fingerprint, iris, palm and voice recognition for 1:1 verification and 1:N identification, with presentation attack detection at every point of capture. We submit our face recognition algorithms to NIST’s Face Recognition Technology Evaluation (FRTE) for independent testing.
Each engine runs the full path from capture to decision, with quality checks, template extraction, comparison and presentation attack detection.
Verification and search on passport, mugshot, selfie and video images, with portrait quality checks before any template is created.
Minutiae and ridge-texture matching for flat, rolled and slap captures, and for contactless capture from a phone camera.
Pupil, limbus and eyelid segmentation, then iris-code comparison built for search across large galleries.
Contactless palmprint and palm-vein recognition from one open-hand capture, aligned on the valleys between the fingers.
Speaker verification on enrolment phrases or free speech, with detection of replayed and synthetic audio.
Passive liveness checks that reject printed photos, screen replays, masks, fake fingers and patterned contact lenses without asking the person to move.
NIST runs each submitted algorithm on sequestered image sets and publishes accuracy, speed and template size for every participant. We submit our face recognition algorithms to NIST’s ongoing Face Recognition Technology Evaluation (FRTE) and list each submission here, with a link to NIST’s report card once results are published.
NIST does not endorse products. Published results describe only the algorithm version NIST tested.
| Programme | What NIST measures | Algorithm | Status |
|---|---|---|---|
| FRTE 1:1 and 1:NFace verification and identification | 1:1 False non-match rate at a false match rate of 10−6 on visa-to-border photos, plus border, kiosk, mugshot and twins image sets.1:N False negative identification rate at fixed false positive identification rates, search time and template size on large galleries. | biometricslive_001 |
Scheduled Q4 2026 |
| PFT IIIFingerprint | False non-match rates for proprietary fingerprint templates across several fingerprint datasets, with template size and comparison speed. | Not yet submitted | Planned Q1 2027 |
| IREX 10 IdentificationIris | False negative identification rate at fixed false positive identification rates on large iris galleries. | Not yet submitted | Planned Q3 2027 |
Hardware H.264 and H.265 decoding of RTSP and ONVIF camera streams.
Each face is followed across frames as a single track.
Every track is scored for pose, blur and lighting, and the clearest frontal frame is kept.
One template per track, compared 1:N against your watchlists.
Matches go to the VMS as ONVIF Profile M events, webhooks or bookmarks.
NVIDIA Jetson Orin modules beside the cameras, sending only events upstream.
GPU servers inside your network, including air-gapped sites.
REST and gRPC endpoints for apps and remote onboarding.
We size edge and server hardware for each site, from a handful of cameras to several hundred.
A C++ core with Python, Java and Kotlin, Swift and C# bindings. Templates and scores match across platforms, so a person enrolled on a phone can be searched on a server.
| Platform | Package | Acceleration |
|---|---|---|
| Linux x86-64 | SDK, Docker image | CUDA, TensorRT |
| Linux ARM64 | SDK | NVIDIA Jetson, TensorRT |
| Windows x64 | SDK | CUDA, ONNX Runtime |
| Android | AAR library | On-device GPU and NPU |
| iOS | XCFramework | Core ML |
import biometricslive as bl
face = bl.FaceEngine(device="cuda:0")
probe = face.template(bl.load_image("probe.jpg"))
reference = face.template(bl.load_image("passport.jpg"))
score = face.compare(probe, reference)
threshold = face.threshold(fmr=1e-6)
print(score >= threshold) # True for the same personA leaked password can be changed. A leaked face or fingerprint cannot, so these are the defaults we build to.
Engines run inside your network. Images and templates leave it only when you send them.
AES-256-GCM encryption at rest and TLS 1.3 in transit, with renewable, unlinkable templates following ISO/IEC 24745.
Consent records, purpose limits and deletion on request, to support obligations under India’s DPDP Act 2023 and the EU GDPR.
False match and false non-match rates measured across age, sex and region-of-birth groups, following ISO/IEC 19795-10.
Tell us the modality, the scale and where it will run. We reply with evaluation SDK access or a hardware sizing call.