1:1 Face Verification & 1:N Face Identification
1:1 Face Verification
Matches a selfie to a single ID photo with 99.99% accuracy, ideal for secure biometric authentication.
1:N Face Identification
Scans vast databases for 1:N face identification, perfect for fraud detection or VIP recognition.
Facial biometrics are used in the KYC process to compare selfie photos with photographs from documents and confirm the presence of a user in biometric “block” and “safe” lists.
Oz Forensics Face Recognition Algorithms allow
Detect and highlight the best shot from the video.
Compare face from photo with face photo from ID.
Perform a search in the biometric database.
Use Cases
Enhance your security and streamline identity verification
Schedule your demo
Ready to enhance security without increasing friction?
Certified to prevent biometric fraud
Our certifications guarantee 100% accuracy in identity verification. See what makes us trusted by global leaders.
What is Oz Biometry and how does face matching work?
Oz Biometry is Oz Forensics’ face matching technology, which confirms that two or more images show the same person. It works by converting each face into a mathematical representation and comparing them, returning a similarity score from 0 (no match) to 1 (or 100%, a full match). It supports both 1:1 verification (is this selfie the same person as this ID photo, for example) and 1:N identification (does this face appear anywhere in a dataset?). Oz’s face matching scored 99.99% accuracy in NIST FRVT.
What’s the difference between 1:1 verification and 1:N identification?
1:1 verification compares one face against one other face to answer “are these the same person?”, for which the classic example is matching a live selfie to the photo on an ID document during onboarding. 1:N identification compares one face against many faces in a database to answer “who is this, or is this person on a list?”, used for watchlist and duplicate-identity checks. Oz Biometry is capable of both 1:1 and 1:N verification.
How accurate is Oz’s face matching?
Oz Biometry scored 99.99% accuracy in NIST FRVT, the most respected independent benchmark for face recognition, which tests algorithms on closed datasets that vendors never see. The model is trained on more than 4.5 million unique faces with diverse ethnic representation and uses additional attributes to keep scores robust across real-world conditions. Accuracy is reported by NIST rather than self-claimed, which is what matters most to buyers evaluating face matching.
What is NIST FRVT and why does it matter for face matching?
NIST FRVT (Face Recognition Vendor Test) is the independent benchmark run by the U.S. National Institute of Standards and Technology that evaluates face recognition algorithms for both 1:1 verification and 1:N identification. It matters because it’s run on closed “black box” datasets that vendors can’t access or train on, so the results can’t be gamed, making it the most credible accuracy signal in the industry. Oz Biometry achieved 99.99% accuracy in NIST FRVT, which is why it’s trusted by banks and government agencies.
Can Oz match a selfie to an ID document or passport photo?
Yes, matching a live selfie to the photo on an ID document or passport is the core 1:1 use case for Oz Biometry, and it’s the heart of remote KYC onboarding. Oz’s face detector is built to handle rotated, compressed, or low-quality document images (the kind you get from a phone photo of a passport), as well as selfies taken in uneven lighting. It returns a similarity score you compare against your threshold to accept or decline the match.
Does face matching work with low-quality or rotated ID photos?
Yes. Oz Biometry’s face detector is specifically designed to cope with the messy reality of ID images, rotated passport scans, compression artifacts, and low-resolution document photos, as well as selfies shot in poor lighting. This robustness matters because the reference photo on an ID is often captured under very different conditions than a live phone selfie, and naive matchers lose accuracy when the two don’t line up. Oz uses additional facial attributes to keep match scores reliable across those conditions.
How does the system handle aging or changes in facial appearance over time?
Oz Biometry is designed to remain reliable as faces change naturally over time, using additional facial attributes and a face representation that focuses on stable, distinguishing features rather than transient details like facial hair, minor cosmetic changes, or weight fluctuations. This matters in practice because the reference photo on an ID document is often years old, so the matching engine has to bridge a real gap between an older photo and a current selfie. Combined with its 99.99% NIST FRVT accuracy, this gives Oz Biometry long-term reliability for use cases like periodic re-verification, not just first-time onboarding.
How does Oz handle a watchlist or blacklist check?
Oz handles watchlist and blacklist screening through its Collection check, which uses 1:N identification to match a person’s face against a pre-existing database of faces. It’s powered by Pgvector (a vector similarity extension for PostgreSQL) and runs an exact nearest-neighbor search — Oz deliberately avoids approximate indexes because they trade away accuracy. A match scoring at or above the threshold is flagged. The whole database is held in memory for performance, and the design works well in scale.
How do you handle database security for large-scale biometric records?
Oz Biometry secures large-scale biometric records with strong encryption, strict access controls, and a database architecture that keeps data protected without sacrificing match speed. The design supports exact nearest-neighbor search at scale, comfortably handling databases at large scales. This is backed by the company’s ISO/IEC 27001:2022 certification and regular SOC 2 Type 2 audits, so enterprise-scale watchlist and de-duplication databases are protected to the same standard as the rest of the platform.
What face-match threshold should I use, and what happens to borderline cases?
The default match threshold in Oz Biometry is 0.85, a similarity score at or above that is treated as the same person, and it’s configurable to your risk tolerance (raise it to be stricter, lower it to be more permissive). For face matching and collection checks, Oz also supports an OPERATOR_REQUIRED outcome: when a result is ambiguous, it can be routed to a human reviewer rather than auto-accepted or auto-rejected. Note that liveness is binary, it only returns SUCCESS or DECLINED, so OPERATOR_REQUIRED applies to face matching and collection, not liveness.
Is Oz’s face matching biased across different ethnicities or demographics?
Demographic fairness depends heavily on training data, and Oz Biometry is trained on more than 4.5 million unique faces selected for diverse ethnic representation, specifically to reduce performance gaps across groups. Its accuracy is validated independently through NIST FRVT (99.99%), which publishes performance across demographic groups rather than relying on a vendor’s own claims. For buyers who care about fairness, independent NIST results plus diverse training data are the two signals to check — and Oz reports both.
What are the technical requirements for integrating the Oz Biometry SDK?
Integrating Oz Biometry requires a standard iOS, Android, Flutter, or Web development environment, and the same SDKs used for Oz Liveness expose 1:1 and 1:N endpoints, so most teams don’t need a separate integration project. Comprehensive documentation covers each API mode: Full API (asynchronous, with storage and webhooks) and Instant API (synchronous, no data stored), letting developers choose the integration pattern that matches their latency, storage, and compliance needs. Oz’s technical support team assists throughout, from initial SDK setup to production-scale tuning.
Can I run face matching without storing biometric data?
Yes, Oz’s Instant API performs a face comparison (or a liveness check) and returns the result immediately while storing nothing, which is built for privacy-first and GDPR-sensitive deployments. If you do need persistence for audit and case management, the Full API stores media and results and provides webhooks. And if data must never leave your environment at all, Oz Biometry can be deployed fully on-premise. The relevant Instant endpoints compare one face to another, or one face against many.
How do liveness and face matching work together in a verification flow?
Liveness and face matching answer two different questions and are strongest together: liveness proves the person is real and present right now, while face matching proves they’re the same person as the ID or the enrolled record. A typical Oz flow runs a liveness check on the live selfie video, then matches the best frame of the video selfie against the ID document photo (1:1) and optionally against a database (1:N).
How is Oz Biometry different from Onfido or Jumio face matching?
Oz Biometry, Onfido, and Jumio all offer selfie-to-document face matching for KYC, so the differences come down to accuracy validation, deployment, and privacy. Oz reports 99.99% accuracy in independent NIST FRVT testing and trains on 4.5M+ diverse faces; it can be deployed on-premise or on-device for data sovereignty rather than cloud-only; and its Instant API lets you match faces without storing any biometric data for GDPR-sensitive use cases. It also pairs natively with Oz Liveness (certified iBeta ISO/IEC 30107-3 Level 1, 2 and 3) and CEN/TS 18099 so liveness and matching come from one certified architecture.
