Deep Learning-Powered ECG-Based Biometric Authentication

A system for biometric authentication utilizes behavioral or biological data to identify individuals. However, some biometric recognition systems, such as which that rely on fingerprints or iris scans, can be circumvented, making them less reliable. To be effective, a biometric system must be both robust and reliable in the face of potential attacks. The electrocardiogram (ECG) is a biological biometric measure that relies on the specific anatomical and geometrical properties of the myocardium in person's heart. It is a unique characteristic that cannot be easily imitated, resulting in highly reliable ECG-based identification that is resistant to fraud and imitation attacks. In fact, ECG-based identification has been shown to be effective in detecting liveness, ensuring permanence, and achieving universality. In this paper, we employ a multiple deep learning based approaches to identify a person using an ECG signal, such as a transformer-based model, long short-term memory(LSTM) based model and Convolutional neural networks(CNNs) based model. We also implement an CNN-based Siamese network for verification of ECG-based biometric dataset. We conducted experiments to assess the effectiveness of our approach using two datasets: the MIT-BIH Arrhythmia and ECG-ID. Our most successful method achieved 98.99% accuracy for identifying individuals in the MIT-BIH dataset and 99.54% accuracy for the ECG-ID dataset. Additionally, for verification purposes, our 2D CNN-based Siamese model obtained an accuracy rate of 86% when applied to the ECG-ID dataset.