Towards Billion-scale Multi-modal Biometric Search

Researchers present Bharat ABIS, a billion-scale multimodal biometric identification system built on open-source architecture that processes fingerprint, face, and iris data for India's Aadhaar database. The system achieves 0.3% false non-match rate on 220 million identities and processes 100 searches per second, demonstrating practical scalability for country-level identity infrastructure.

Bharat ABIS represents a significant engineering achievement in deploying biometric systems at unprecedented scale. The project tackles fundamental infrastructure challenges that emerge when matching individuals across 1.5+ billion records, requiring innovations across the entire pipeline from data acquisition to matching algorithms. This work is particularly relevant given the critical role identity systems play in financial services, government administration, and security operations across developing nations.

The system's architecture combines fingerprint, facial, and iris recognition into a unified matching framework, with each modality contributing to accuracy while managing the computational complexity of billion-scale searches. By achieving sub-0.3% false rejection rates at reasonable false acceptance thresholds on demographically diverse populations, Bharat ABIS demonstrates that open-source approaches can match proprietary commercial systems in real-world scenarios. The 13.5KB per-person template size and 100-search-per-second throughput on modest hardware (single server with 8 H100 GPUs) indicates efficient feature extraction and indexing strategies.

For the identity infrastructure and government technology sectors, this research validates that large-scale biometric systems can be built without complete dependency on Western commercial vendors. The technical insights around handling special cases (missing digits), presentation attack detection, and quality assessment across diverse populations provide practical blueprints for other nations implementing similar systems. The open-source foundation enables knowledge transfer and reduces vendor lock-in risks. However, the system's effectiveness depends heavily on data quality, demographic representation, and security protocols that aren't detailed in this academic presentation. Organizations implementing such systems must address privacy considerations, data governance, and security safeguards beyond the technical accuracy metrics presented.

• →Bharat ABIS achieves 0.3% false rejection rate across 220 million diverse identities, comparable to state-of-the-art commercial systems
• →System processes 100 searches per second on standard GPU hardware, proving billion-scale biometric search is computationally feasible
• →Multimodal approach combining fingerprint, face, and iris recognition improves accuracy and robustness compared to single-modality systems
• →Open-source architecture reduces vendor dependency for large-scale identity infrastructure projects in developing nations
• →Technical handling of edge cases (missing digits, presentation attacks, quality assessment) enables real-world deployment challenges