A multimodal biometric recognition system based on Fingerprints, Iris and ECG via Swin Transformer and CNN Model

Biometric recognition systems have become a vital component of modern security and identity management infrastructures. Multimodal biometric systems integrate multiple biometric traits for identification or verification, offer notable advantages over unimodal biometric systems such as improved accuracy, enhanced security, and greater robustness. Among individual modalities, fingerprint recognition is widely adopted for its reliability and ease of use; iris recognition is renowned for its high accuracy and stability; and electrocardiogram (ECG) signals present a unique, behaviour-based biometric trait that is difficult to replicate. This study highlights the importance of live biometric ECG signals to enhance the security of existing multimodal recognition systems against spoofing attacks. It introduces a novel feature extraction approach that simultaneously processes spatial (position-based) features and attention-based features in parallel. This work also proposes a novel fusion of fingerprint, iris, and ECG modalities to exploit their complementary strengths and mitigate spoofing risks. A convolutional Swin Transformer architecture is introduced for effective feature extraction from each modality. For evaluation, a custom multimodal biometric dataset was created by combining the IITD iris database, the HEARTPRINT ECG dataset, and the SOCOfing fingerprint dataset. The proposed model was first applied individually to each modality, achieving recognition accuracies of 96 % for fingerprints, 97 % for iris, and 71 % for ECG. Subsequently, the model was evaluated on the fused multimodal dataset, yielding a recognition accuracy of 99 %. A p-value of 0.01 from the Chi-Square test provides strong evidence that the model's performance is statistically significant. These results underscore the robustness and effectiveness of the proposed system in resisting spoofing attacks and ensuring secure biometric recognition. This study contributes to the advancement of multimodal biometric systems by demonstrating the efficacy of advanced feature extraction techniques and robust fusion strategies.