Enhanced biometric template protection schemes using distance based fuzzy extractor

Abstract

The biometric template protection systems are essential for improving the security of biometric authentication systems in Internet of Things (IoT)-based applications. However, insufficient user data, compromised keys, and privacy concerns raise significant challenges regarding the reliability and security of these systems. One critical challenge is unauthorized access to biometric templates, which exposes users to potential security threats. The proposed system addresses this by employing a novel technique that enhances template security through a cancellable biometric (CB) scheme. While CB schemes improve security by applying a one-way transformation to the biometric template, they often suffer from decreased accuracy due to the complexity of transformations applied to the feature vector. To overcome these limitations, the proposed system integrates a Self-learning based Multi-scale Residual Convolutional Neural Network (SM-ResCNN) for feature extraction, which improves classification accuracy by capturing features at various scales. These features are then classified by an Enhanced Random Forest (MRF) classifier, ensuring high accuracy while mitigating overfitting. Additionally, the Distance-based Fuzzy Extractor (DFE) is employed for cancellable template protection, converting biometric data into uniformly arbitrary and reproducible random strings, enhancing security without compromising performance. The performance of the proposed approach is simulated in the FERT and CASIA datasets and contrasted with state-of-the-art methods. The recognition rates obtained with the FERET and CASIA datasets are 99.81 % with 0.015 Equal error rate (EER) and 99.7 % with 0.0211 EER, respectively. The study shows that the proposed method significantly improves biometric authentication security while maintaining high classification accuracy, outperforming existing state-of-the-art methods.