Multimodal Biometric Template Protection Based on Deep Binary Encoding

The adoption of multimodal biometric systems in forensic investigations is accelerating. While these systems offer high verification accuracy, the extracted features stored in unsecured raw form in forensic databases pose significant privacy risks. This work proposes a cancelable multimodal template protection method using deep feature extraction. The extracted binarized features are fused and transformed into secured discretized vectors represented as irreversible binary strings. The scheme ensures template diversity, revocability, and resistance to reconstruction attacks, adhering to ISO/IEC 24745:2022. Experiments on fingerprint (FVC2002-DB1), face (LFW), and iris (IITD-Iris) datasets show strong performance and security over existing methods.