Off-Person ECG Biometrics Using Spatial Representations and Convolutional Neural Networks

Iulian B Ciocoiu,Nicolae Cleju

doi:10.1109/access.2020.3042547

Iulian B Ciocoiu, Nicolae Cleju

Open Access

PDF Available

https://doi.org/10.1109/access.2020.3042547

Copy DOI

Export

Save

Cite

Journal: IEEE Access	Publication Date: Jan 1, 2020
Citations: 9	License type: CC BY 4.0

Affiliation: Gheorghe Asachi Technical University of Iași

Abstract
Highlights/Summary
Full-Text PDF
Similar Papers

Abstract

Listen

Targeting off-person ECG-based biometrics, we report a comparative analysis of identification accuracy and verification Equal Error Rate (EER) performances of four distinct types of spatial representations of ECG signals applied as inputs to Convolutional Neural Networks. The actual algorithms used to transform the original time series into 2D/3D images are based on a modified version of the Continuous Wavelet Transform (the S-Transform), the Gramian Angular Field, the recurrence plot, and state-space representations. Extensive experiments have been conducted using UofT and CYBHI datasets including recordings acquired on fingers and hand palm under various activity scenarios. The wavelet-based approach yielded best results, while all analyzed solutions compare favorably with previously reported performances.

Highlights

During the last decades the biometrics landscape has been largely dominated by systems which acquire and process fingerprint, face, iris, speech data or combinations of those
ECG signals have been mostly preferred for biometric applications, due to their inherent liveness properties, difficulties associated with fraudulent recording, and increased availability of affordable, wearable, user-accepted sensory acquisition devices
We describe the datasets used in the experiments, the preprocessing procedures, the CNN architecture, and provide comparative identification accuracy and verification Equal Error Rate (EER) performances of the analyzed approaches against existing solutions

Summary

Introduction

During the last decades the biometrics landscape has been largely dominated by systems which acquire and process fingerprint, face, iris, speech data or combinations of those. Due to the increasing success rate of counterfeiting or circumventing such approaches, researchers have been forced to consider alternative sensory information and better decision-making solutions, while keeping a comfortable, user-friendly acquisition and operation setup. ECG signals have been mostly preferred for biometric applications, due to their inherent liveness properties, difficulties associated with fraudulent recording, and increased availability of affordable, wearable, user-accepted sensory acquisition devices. While still facing problems related to variability induced by human activity, depreciation of signal-to-noise ratio in case of off-person acquisition, or time lapse between training and testing sessions, ECG-based biometrics has been considered by a significant number of studies that have been extensively surveyed recently [4]. There is a clear need for additional studies using off-person acquired data, including public availability of reliable, long-term, activitydependent recordings for a significant number of people. Despite practical appeal, experiments conducted on ECG records acquired off the person are rather scarce, and clearly outnumbered by those using sensors placed on the chest

Results

Discussion

Conclusion