Analyzing personalized policies for online biometric verification.

Area of Science:

• Biometrics and Identity Management
• Machine Learning and Pattern Recognition
• Societal Impact of Technology

Background:

• India's national biometric program aims for social inclusion using unique individual identification.
• Biometric verification relies on matching current data against enrolled templates.
• Existing verification methods often involve multiple biometric traits like fingerprints and irises.

Purpose of the Study:

• To develop and evaluate individualized biometric verification strategies.
• To optimize verification processes by minimizing false reject rates (FRR) and mean verification delay.
• To maintain acceptable false accept rates (FAR) within a large-scale national program.

Main Methods:

• Utilized similarity scores from 10 fingerprints and 2 irises for individualized strategies.
• Developed a probabilistic model for the joint distribution of 12 biometric similarity scores.
• Implemented a likelihood ratio test for genuine vs. imposter classification.
• Considered two-stage policies for inconclusive first-stage results.

Main Results:

• Individualized strategies achieved the same FRR as acquiring all 12 biometrics.
• Demonstrated significant FRR reduction compared to previous fingerprint and iris-only policies.
• Achieved a mean verification delay of [Formula: see text] seconds, substantially lower than multi-modal approaches.
• Optimized policies required iris scans for 32-41% of residents, averaging 1.3 fingerprints per verification.

Conclusions:

• Individualized biometric verification strategies offer significant efficiency gains.
• The proposed method balances accuracy (low FRR/FAR) with speed (reduced delay).
• This approach enhances the feasibility and effectiveness of large-scale biometric identification systems.