Open Access Open Access  Restricted Access Subscription Access

A Hybrid Deep Learning Framework for Multimodal Biometric Verification Utilizing ECG, Fingerprint, and Iris Data

B. Naga Mallesh, Dr. T. Raghunadha Reddy, K.V. Sai Jayanth, D. Raju

Abstract


Single-trait biometric systems frequently face criti- cal security flaws, ranging from physical sensor wear to sophisti- cated spoofing attacks. When deploying modern deep learning to solve these issues, developers encounter a new problem: data starvation. Standard classification models require thou- sands of images to generalize, leading to massive overfitting when only a few enrollment scans are available. To overcome this limitation, we engineered a robust multimodal biometric architecture tailored specifically for data-scarce constraints (a maximum of three training samples per user). Our solution fuses a dynamic, one-dimensional behavioral signal (Electrocar- diogram) with static, two-dimensional physical traits (Finger- print and Iris) across a dataset of 199 unique individuals. To bypass the failure of standard Convolutional Neural Networks (CNNs) on sparse physical data, we shifted the network objective from 1-to-N identification to 1-to-1 verification using Siamese architectures. We processed the physical traits through parallel Siamese ResNet50V2 networks optimized with contrastive loss to generate 512-dimensional feature embeddings. Simultaneously, a custom-built 1D-CNN evaluated the temporal ECG data. We then normalized and merged the outputs from these varying modalities using a targeted, weighted score-level fusion algorithm. Testing showed that while traditional 2D Softmax classifiers stalled below 40% accuracy, our Siamese metric learning approach drastically elevated unimodal performance. The combined multimodal sys- tem (anchored by a 50% ECG, 30% Iris, and 20% Fingerprint weight distribution) achieved a final accuracy of 92.21%. Further- more, the system recorded a False Acceptance Rate (FAR) of just 1.01% alongside a 14.57% False Rejection Rate (FRR), proving that our hybrid fusion methodology mathematically neutralizes the weaknesses of individual biometrics in highly constrained environments.


Full Text:

PDF

References


A. K. Jain, A. Ross, and S. Prabhakar, “An introduction to biometric recognition,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 14, no. 1, pp. 4-20, 2004.

L. Biel, O. Pettersson, L. Philipson, and P. Wide, “ECG analysis: A new approach in human identification,” IEEE Transactions on Instrumenta- tion and Measurement, vol. 50, no. 3, pp. 808-812, 2001.

S. Silva, et al., “Using Convolutional Neural Network and a Single Heartbeat for ECG Biometric Recognition,” Entropy, vol. 23, no. 6,p. 733, 2021.

I. Almakdi, et al., “Electrocardiogram (ECG)-Based User Authentication Using Deep Learning Algorithms,” Healthcare, 2024.

M. S. Rahman, et al., “Deep Learning-Powered ECG-Based Biometric Authentication,” IEEE International Conference on Networking, Com- munication and Information Management (NCIM), 2023.

J. Daugman, “How iris recognition works,” IEEE Transactions on Circuits and Systems for Video Technology, vol. 14, no. 1, pp. 21-30, 2004.

Y. Taigman, M. Yang, M. Ranzato, and L. Wolf, “DeepFace: Closing the gap to human-level performance in face verification,” in Proceedings of the IEEE CVPR, pp. 1701-1708, 2014.

J. Bromley, I. Guyon, Y. LeCun, E. Sa¨ckinger, and R. Shah, “Signature verification using a’siamese’ time delay neural network,” Advances in neural information processing systems, vol. 6, 1993.

R. Hadsell, S. Chopra, and Y. LeCun, “Dimensionality reduction by learning an invariant mapping,” in Proceedings of the IEEE CVPR, vol. 2, pp. 1735-1742, 2006.

X. Wang, et al., “Multi-Task Deep Learning for Multimodal Biometric Recognition,” IEEE Access, 2023.


Refbacks

  • There are currently no refbacks.