Iris Identification at National Scale
According to the World Bank, around 850 million people have no official proof of identity, and most countries closing this gap build national registers with biometrics. Such a register has one biometric question that matters more than any other: is this person already registered under another name? Iris answers it better than most biometric modalities. The harder problem is that the answer often comes together with a vendor lock that lasts for decades. This is the reason we started building open-source biometric components.
Why iris for population registers
De-duplication is 1:N search at its most demanding. Every new enrollment is compared with everyone enrolled before. With tens of millions of records, even a tiny false match rate produces a steady flow of false alarms, and each of them becomes a case for a human adjudicator. Iris codes separate different people with an unusually wide margin, which allows a program to set a strict threshold without rejecting genuine applicants. This is why India's Aadhaar enrolls both irises next to ten fingerprints, and why many newer programs use iris where fingerprints are worn by manual work or age.
Iris also helps in the field. It is contactless, a handheld scanner works with a laptop or a tablet, and the capture takes seconds. For registration drives in rural areas, in refugee settings or for door-to-door surveys for social benefits, this matters as much as accuracy.
The cost that does not show up in the tender
A typical biometric system is bought as a package: scanners, enrollment software, an automated biometric identification system (ABIS) and the vendor's template format. The price looks fine. The problems start years later:
- Templates are stored in the vendor's own format. Switching the matcher means re-processing every original image, if the images were kept, or re-enrolling the population, if they were not.
- Scanners are certified only for this vendor's software. Replacing broken devices means buying from the same supplier at the supplier's price.
- Licenses are counted per record. The register grows every year, and so does the bill.
The technology itself may be good. The problem is who controls the data: if a country cannot move its population register to another system without the vendor's help, it does not fully own it.
What an open setup looks like
Standard formats everywhere. Iris records in ISO/IEC 19794-6 or 39794-6, fingerprint templates in ISO/IEC 19794-2 or ANSI INCITS 378, faces in ISO/IEC 19794-5. Anything stored should be readable by more than one product. Where legacy data is in another format, convert it once and keep the originals. Our open-source Biometric Converter does this for fingerprint templates, and you can try it in the browser.
Open platform, replaceable components. MOSIP, the open-source identity platform used by more than a dozen countries, is built this way. The platform defines interfaces for biometric SDKs and ABIS, and the country plugs in the implementation it chooses. This way a component that does not perform can be replaced without rebuilding the register.
Open-source engines with paid support. The matching and conversion code at the heart of the register should be possible to inspect and run without a license server. Support, integration, tuning and certification are where a vendor should earn money, not in keeping the customer locked in.
What we can do for your program
- Integrate with MOSIP or your existing platform. Our face match SDK is already used in Inji Wallet, the reference wallet of MOSIP, to check a selfie against the photo in a credential right on the phone.
- Migrate a legacy biometric database to ISO and ANSI formats, so the register is no longer tied to one vendor.
- Set up enrollment in the field with iris scanners and the Edge service, which works offline and synchronizes when the connection is back.
- Adapt the models to your population: retrain them on representative local data and tune the thresholds for your target false match rate.
- Support the system for the long term on open-source components, which you can inspect and keep running even without us.