A dual-architecture approach to ECG biometric verification and authentication

Abstract

A dual-architecture approach to ECG biometric verification and authentication

Azab M.A., Korzhuk V.M.

Incoming article date: 23.03.2025

This paper presents a highly technical implementation of an ECG-based biometric identification system utilizing deep learning models for both verification and closed-set identification. We propose a dual-model architecture comprising a Siamese neural network for one-to-one verification and a deep convolutional neural network (CNN) for one-to-many classification. The methodology includes comprehensive signal preprocessing, data augmentation to simulate physiological variability, and feature extraction tailored to ECG characteristics. Experimental evaluation on benchmark ECG datasets demonstrates the effectiveness of the proposed system. The Siamese network achieves high verification accuracy with low equal error rates, while the CNN classifier attains state-of-the-art identification accuracy (exceeding 98% on average) across enrolled subjects. Key performance metrics—accuracy, precision, recall, and F1-score—indicate robust performance, outperforming several existing biometric methods. The results highlight the viability of ECG-based authentication in real-world applications.

Keywords: biometric authentication, electrocardiogram (ECG), siamese neural network, convolutional neural network, qrs complex, signal processing