Electrocardiograms (ECGs) have shown unique patterns to distinguish between
different subjects and present important advantages compared to other biometric
traits, such as difficulty to counterfeit, liveness detection, and ubiquity.
Also, with the success of Deep Learning technologies, ECG biometric recognition
has received increasing interest in recent years. However, it is not easy to
evaluate the improvements of novel ECG proposed methods, mainly due to the lack
of public data and standard experimental protocols. In this study, we perform
extensive analysis and comparison of different scenarios in ECG biometric
recognition. Both verification and identification tasks are investigated, as
well as single- and multi-session scenarios. Finally, we also perform single-
and multi-lead ECG experiments, considering traditional scenarios using
electrodes in the chest and limbs and current user-friendly wearable devices.
In addition, we present ECGXtractor, a robust Deep Learning technology
trained with an in-house large-scale database and able to operate successfully
across various scenarios and multiple databases. We introduce our proposed
feature extractor, trained with multiple sinus-rhythm heartbeats belonging to
55,967 subjects, and provide a general public benchmark evaluation with
detailed experimental protocol. We evaluate the system performance over four
different databases: i) our in-house database, ii) PTB, iii) ECG-ID, and iv)
CYBHi. With the widely used PTB database, we achieve Equal Error Rates of 0.14%
and 2.06% in verification, and accuracies of 100% and 96.46% in identification,
respectively in single- and multi-session analysis. We release the source code,
experimental protocol details, and pre-trained models in GitHub to advance in
the field.

本研究利用一个大型数据库训练出深度学习技术的特征提取器，结合多种单、多阶段及多种导联心电图实验，实现了 ECG 生物识别技术的精细分析及比较，并通过多个公共数据库的验证，进一步优化了模型，提出了行之有效的改进建议。