Genetic algorithms and deep learning for unique facial landmark-based key generation

Generating secure secret keys remains essential in the realm of biometric authentication systems. Traditional methods have often suffered from inefficiency, insecurity, or the requirement for additional hardware. In this study, an innovative approach to secret key generation is proposed, leveraging deep learning, facial feature extraction, genetic algorithms, and linear feedback shift registers (LFSRs). These techniques are combined to create robust, unique keys based on users' facial features. A convolutional neural network (CNN) is employed for the extraction of facial features from user images and for the optimization of LFSR parameters using a genetic algorithm. Furthermore, another neural network is utilized to establish a connection between facial features and the LFSR-generated output, resulting in the secret key. Rigorous evaluation of the method is conducted across various facial datasets, with a comparison against existing approaches. The results demonstrate the effectiveness of the method, yielding secret key length of 92,706 characterized by an entropy value of 3.24, a low average correlation value of 0.1554, and a high level of security. This research represents a significant advancement in secure biometric authentication, addressing the limitations of conventional key generation methods.