Improved biometric data protection: Bounded brute-force strategy for maximum likelihood decoding

Conventional biometric data protection schemes often struggle to provide strong and reliable security guarantees after transformation, largely due to the noise amplification introduced during quantization. This amplified noise can distort the relationship between the protected and original biometric data, creating a gap between the claimed security of the protected representation and the actual security of the raw input. Such a mismatch risks overestimating system robustness and may expose the scheme to vulnerabilities such as pre-image attacks. To address this challenge, we propose a novel secure sketch construction that integrates Locality-Sensitive Hashing (LSH) with a bounded brute-force strategy for maximum likelihood decoding. Our method achieves asymptotically optimal error tolerance while preserving the statistical alignment of inter- and intra-class variability across both unprotected and protected domains. This alignment enables accurate key recovery and enhances resistance to pre-image and decoding attacks. Comprehensive experiments demonstrate that our method consistently outperforms existing approaches in both security and robustness to biometric variability, offering a practical and theoretically grounded solution for biometric authentication.