Secure cancelable multimodal biometrics using hashing - based random projection

Multimodal biometric systems improve recognition accuracy and reduce vulnerabilities common in unimodal systems. Biometric template security faces a significant challenge because compromised templates cannot be reused like passwords. The proposed system addresses the risk of secure and revocable protection of biometric templates while maintaining high authentication performance. It introduces a cancelable multimodal biometric system that incorporates both face and fingerprint traits. In the given system, discriminative features are extracted in both modalities, and feature-level fusion is carried out in order to form a binary representation. SHEP (Secure Hashing, Exponent Warping, and Random Projections) converts the raw features to a non-invertible and revocable representation, and it is presented as a cancelable template generation method that improves the security of the fused template. The face and fingerprint dataset used to test our approach include CASIA-FaceV5 and CASIA-FingerprintV5. The Equal Error Rate (EER) and Receiver Operating Characteristic (ROC) curves are used to evaluate the performance. The experimental findings indicate that a multimodal SHEP-based cancelable template with a 256-dimensional fused cancelable template, EER=0.000, ROC(AUC) = 1.000 and verification rate=100% demonstrates that SHEP is very effective in maintaining accuracy as well as security.