Deep Learning-Based Multimodal Biometric System: A Fusion Approach Integrating Iris, Face, and Finger Vein Traits

In the current era of ubiquitous data and connectivity, information is constantly shared across networks, devices, and databases, driving innovation and progress across various sectors. However, this increase in connectivity also poses significant risks to the privacy and confidentiality of individuals and organizations. Traditional security measures like passwords, cards, and PINs are increasingly inadequate, leading us to explore more robust techniques for safeguarding our data and access controls. Biometric authentication emerges as a powerful solution. It can uniquely identify individuals based on their physiological or behavioral traits and is widely accepted by users. Note that to avoid various issues and identity theft attempts, it is better to use more than one trait, prompting the shift from unimodal to multimodal biometric systems. This work presents a deep learning-based multimodal biometric system that integrates iris, face, and finger vein recognition. The proposed approach utilizes advanced preprocessing and region of interest (ROI) extraction techniques, including the Segment Anything Model (SAM) and MediaPipe. Feature extraction is performed using pre-trained convolutional neural networks (CNNs) such as ResNet and FaceNet, as well as transformer architectures like Vision Transformer (ViT). A score-level fusion strategy allows for flexible use of individual unimodal systems or their combination to improve overall performance. The system is trained and evaluated on a dataset containing real-world biometric samples from all three modalities. Experimental results show a recognition accuracy of 99%, demonstrating the effectiveness of combining multiple biometric traits with deep learning techniques for secure and accurate authentication.