Open-source-based Bharat ABIS shows potential for national-scale deduplication

A team of researchers mostly affiliated with the Unique Identification Authority of India (UIDAI) has custom-built a multimodal biometric search system to perform effective and cost-efficient deduplication at national scale.

The six UIDAI researchers, along with a researcher from IIIT Hyderabad and Anil Jain of Michigan State University, published their findings in the paper “Towards Billion-Scale Multi-modal Biometric Search.”

The researchers used 220 million demographically-controlled identity records randomly sampled from Aadhaar’s 1.55 billion records for the test. The Bharat ABIS they developed combines fingerprint, iris and face biometrics from each record into a single concatenated template 13.5KB in size.

Bharat was built with an open-source architecture, the researchers say. The model puts each biometric modality through a series of preprocessing and processing steps, including segmentation, quality assessment and presentation attack detection (PAD). A similarity search module compares the probe to all records in the gallery and a scheduler distributes and coordinates tasks across different servers to handle scale. The modalities are weighted for the comparison process.

The ABIS delivered a false negative identification rate (FNIR) of 0.3 percent at a false positive identification rate (FPIR) of 0.5 percent with adults’ records. On a gallery of 20 million records, it delivered an FPIR of 0.1 percent at an FNIR of 0.05 percent, which the paper notes exceeds Aadhaar’s design requirements.

This level of accuracy significantly surpasses the recorded performance of single-modality deduplication.

The researchers also benchmarked their ABIS against the three unidentified commercial off-the-shelf (COTS) ABIS platforms used in the Aadhaar system. The results showed the Bharat ABIS’ performance as “comparable or better,” according to the researchers.

Aadhaar has used commercial ABIS from developers including Neurotechnology and the predecessors of Idemia and Thales, three of the world’s leading biometric algorithm providers.

This is important, the researchers say, because many countries cannot afford an off-the-shelf ABIS solution, and have populations of 20 million or fewer. This makes the open-source development approach the researchers used to build their Bharat ABIS a viable option, they say, for all of those countries.

Article Topics

ABIS - Automated Biometric Identification System  |  Anil Jain  |  biometrics  |  biometrics research  |  deduplication  |  India  |  open source  |  UIDAI