FINGER VEIN-BASED BIOMETRIC KEY GENERATION WITH POST-QUANTUM ENCRYPTION AND OPTIMIZED MAGNETIZED HOPFIELD GLOBAL NEURAL NETWORK

Biometric authentication, especially using finger vein patterns, offers a reliable and non-invasive method for secure identity verification. However, current systems face limitations in robustness, scalability, and post-quantum resilience. Existing methods often struggle with noisy biometric inputs, key instability, and vulnerability to quantum attacks. This motivates the development of a lightweight, secure, and stable key generation system. This paper introduces a new biometric key generation framework based on the combination of several state-of-the-art techniques. It starts with Adaptive and Propagated Mesh Filtering (APMF) for improved preprocessing, followed by Spike-Driven Transformer (SDT) for spatiotemporal feature extraction. Min–Max Normalization (M2N) normalizes feature vectors, which are then mapped to binary strings for robust key representation. A fuzzy extractor produces helper data, and Dynamic Lightweight Symmetric Encryption (DLSE) provides post-quantum secure binding. Authentication and key recovery are carried out through a Magnetized Hopfield Global Neural Network (MHGNN), the weight parameters of which are optimized by the White Shark Optimizer (WSO). The model attains 99.7% accuracy with an Equal Error Rate of 0.08, which is higher compared to traditional methods with regard to accuracy and security. This paradigm provides secure authentication with minimal computational expense, and future extensions will be explored to extend it to multimodal biometrics and real-time deployment on edge devices.