CBSB: Robust Cancelable Biometric System for Banks Using Deep Learning
Bassel Yasser Sayed, Amel Mohamed Mahmoud, El‐Sayed M. El‐Rabaie, Nariman Abdel-Salam · 2023
Biometric authentication is a reliable and convenient technique for confirming a person's identity by analyzing their distinct physiological or behavioral traits. While biometric authentication offers several benefits compared to conventional methods, it confronts challenges related to privacy and security. To address these concerns, cancelable biometric systems (CBS) convert biometric data into non-invertible templates, allowing for revocation and re-issuance if security is compromised. CBS also enhances diversity and renewability in authentication solutions. In this paper, we introduce CSBS: A robust deep-learning-based framework for cancelable biometric systems for banks. The system extracts distinct characteristics from fingerprint and facial images and subsequently transforms them into revocable templates through the Format-Preserving Encryption (FPE) methodology. These consolidated templates are securely archived within a database server, employing the Bloom filter data structure, and we compare them during the biometric authentication process using the Cosine similarity measure. We apply different deep learning models, such as VGG-16 and LeNet-LSTM, to extract features from the biometric images and achieve the best accuracy rates of 98.85% for face and 99.36% for fingerprint recognition respectively. The experimental results show that the system can generate efficient cancelable biometric templates that can be used for banking authentication without compromising the original biometric data.