BIO-FuseNet: A Secure Biometric Fusion Network for Iris and Face Recognition

Abstract

Biometric authentication plays a vital role in protecting sensitive data; however, traditional mechanisms such as passwords and tokens remain susceptible to loss, theft, and misuse. Although unimodal biometric systems are limited by poor data quality and higher error rates, multimodal biometric approaches offer improved robustness and reliability. This work proposes a novel secure multimodal biometric fusion framework that integrates facial and iris recognition using Convolutional Neural Networks (CNNs) combined with variance-based discriminative feature selection and cryptographic template protection. Unlike conventional fusion-based systems that treat security as a post-processing step, the pro- posed framework embeds security directly into the fusion pipeline, ensuring template irreversibility, unlink ability, and resistance to cross-matching attacks without compromising recognition performance. Multiple fusion strategies were systematically evaluated, including feature-level, decision-level, score-level, and a newly designed enhanced score-level fusion mechanism. Experimental results demonstrate that the proposed fusion strategy consistently outperforms existing methods, achieving an accuracy of 97.5 % and a low Equal Error Rate (EER) of 0.25%, which exceeds state-of-the-art multimodal biometric systems. Extensive experiments conducted on the labelled Faces in the Wild (LFW) and Chinese Academy of Sciences Institute of Automation (CASIA-iris) benchmark datasets validate the effectiveness, security, and practical applicability of the proposed framework for high-security and real world authentication scenarios, such as smart infrastructure and access-controlled environments.