Vulnerability Analysis of Low-Entropy Face Recognition Systems against Brute Force Attacks

This research investigates the security vulnerabilities of low-entropy-based facial recognition systems, emphasizing susceptibility to brute-force attacks. Utilizing the Haifa University face databases, Principal Component Analysis (PCA), and the "inverse sampling" method, our study meticulously trains models and synthesizes artificial faces. A Python-based brute-force attack script, executed on a testing database using DLib representation, provides detailed statistics. Empirical experimentation establishes an optimal face-matching threshold of 0.4. System scripts offer flexibility in loading databases, a time-based recovery mode, and visualizing synthesized faces. Analysis reveals the compromise of system security due to low entropy, underscored by hypothetical case studies. Future prospects include advanced countermeasures, dynamic threshold adaptations, ethical considerations, user-centric security measures, continual threshold evaluations, multimodal integration, human-centric design, and continuous monitoring protocols. These avenues aim to fortify system security, addressing emerging challenges and ensuring the reliability of facial recognition in diverse contexts.