Revisiting security: An experimental attack analysis of bio-hashing through the random projection matrix

Biometric authentication is a prominent method for granting access to smart buildings through fingerprints or facial recognition; however, these systems exhibit significant security deficiencies, including privacy violations, biometric data reconstruction, and unauthorized access via spoofing attacks. To address these issues, bio-hashing is used to store biometric data in an encrypted hash template. Each time a user wants to authenticate, the biometric characteristics are extracted. A new hash template has been constructed, which must be compared to the stored template. This study evaluates bio-hashing against the similarity attacks with PSO and CMA-ES optimization. The results indicate that bio-hashing delivered 7% security against similarity attacks with PSO optimization and was highly successful against similarity attacks with CMA-ES optimization (25%). This work contributes to the identification of security inefficiencies associated with the bio-hashing approach by generating adversarial samples with a similarity rate of 93% and 75%, respectively, using two distinct optimizations, PSO and CMA-ES.