Researchers have unveiled Bharat ABIS, an open-source system designed to improve large-scale biometric identification in India’s Aadhaar programme, promising lower error rates and better performance for the country’s vast biometric database.
India’s biometric identity system is under pressure from scale, and researchers linked to the Unique Identification Authority of India have unveiled an open-source tool they say could ease the burden. Bharat ABIS, short for Automated Biometric Identification System, is designed to reduce duplicate enrolments by matching fingerprints, iris scans and facial data against a very large database, while keeping search times and error rates low.
The project brings together six UIDAI researchers, a researcher from IIIT Hyderabad and Anil Jain, the biometrics specialist from Michigan State University, according to the study, “Towards Billion-Scale Multi-modal Biometric Search”. India Today reported that the team tested the system against a gallery of 220 million Aadhaar records, using de-identified data, even though the model itself was trained on a much smaller set of between 100,000 and 200,000 entries.
Bharat ABIS combines three biometric traits into a single 13.5KB template, which is then checked for possible matches in the database. The researchers say the system uses a scheduler to spread searches across servers, a feature intended to keep performance stable as records are added. India Today said the platform can also work with one biometric trait at a time, but that accuracy improves significantly when all three are used together.
On the performance numbers highlighted in the study, Bharat ABIS posted a false positive identification rate of 0.1% and a false negative identification rate of 0.05% on a 20 million-record database, both below the targets set by UIDAI for vendors. The article also said the system matched or outperformed three existing tools used within Aadhaar on that database. In a further claim from the researchers, pairing the system’s DeepPrint fingerprint model with vendor tools could cut authentication failures by half, with the biggest gains seen among older people and manual labourers whose fingerprints are more worn.
The study also suggests the system can handle 100 searches a second on a 40 million-record gallery using a single server with eight Nvidia H100 GPUs and 2TB of memory. Because the code is open source, the researchers argue that it could be useful beyond India, particularly for countries with populations under 20 million that may not be able to afford proprietary biometric systems. Even so, the paper stops short of calling it a finished answer for Aadhaar’s needs, saying Bharat ABIS still has to be tested more extensively at scale before firm conclusions can be drawn.
Disclaimer: This article is intended to inform and educate, not to recommend or endorse any financial product, investment or strategy. Please consider your own financial circumstances and seek professional advice where appropriate before making financial decisions.
