A post-quantum privacy-preserving multimodal biometric authentication framework integrating ML-KEM (Kyber) and fully homomorphic encryption

Abstract

Large-scale quantum computing poses a significant threat to classical public-key cryptography, as Shor’s algorithm can break widely used schemes such as RSA and elliptic curve cryptography, while Grover’s algorithm reduces the effective security of symmetric key systems. In parallel, deepfake-based spoofing attacks present serious challenges to unimodal biometric authentication. However, existing solutions often address post-quantum security, privacy preservation, and multimodal biometric fusion separately. This paper proposes a unified post-quantum, privacy-preserving multimodal biometric authentication framework that integrates ML-KEM (Kyber-1024), a lattice-based post-quantum key encapsulation mechanism, with Cheon–Kim–Kim–Song (CKKS)-based fully homomorphic encryption for secure encrypted-domain biometric matching. Multimodal biometric features (face, iris, and fingerprint) are fused and optimized using a genetic algorithm-based feature selection mechanism, while similarity computation is performed directly on encrypted biometric representations. The proposed framework ensures that biometric templates remain protected during both storage and processing, while secure session keys are established using quantum-resistant mechanisms over authenticated communication channels. Experimental evaluation using public PolyU biometric datasets demonstrated no observed false acceptances across 250,000 impostor authentication trials, corresponding to a low empirical False Acceptance Rate upper bound under 95% confidence estimation, alongside low false rejection rates and scalability for up to 5,000 users. By leveraging SIMD batching and parallelized encrypted processing, the framework achieved amortized end-to-end authentication latency of approximately 65–75 ms per query under optimized batched execution while maintaining robustness under biometric noise perturbations of up to 2% without significant performance degradation. Although fully homomorphic encryption introduces additional computational overhead, the proposed framework achieves a practical balance between strong privacy guarantees and operational efficiency. Overall, this study presents a unified architecture for quantum-resilient and privacy-preserving biometric authentication capable of addressing both cryptographic and AI-driven threats in modern security environments.