Finger ECG based Two-phase Authentication Using 1D Convolutional Neural Networks.

Area of Science:

• Biometrics
• Machine Learning
• Signal Processing

Background:

• Electrocardiogram (ECG) signals offer unique physiological characteristics for user authentication.
• Traditional authentication methods face challenges with security and user convenience.
• Developing robust and efficient biometric systems using physiological signals is an active research area.

Purpose of the Study:

• To propose and evaluate a novel two-phase 1D convolutional neural network (CNN) approach for ECG-based authentication.
• To assess the system's performance using finger-lead ECG signals collected via a mobile device.
• To demonstrate the system's efficiency and high specificity in identifying individuals.

Main Methods:

• A two-phase classification strategy employing two types of CNNs: a general CNN for preliminary screening and a person-specific CNN for fine-grained identification.
• Feature learning and classification are automated within a single CNN structure.
• Utilizing finger ECG signals acquired across different sessions and varying sample sizes.

Main Results:

• The proposed two-phase CNN system achieved a promising Equal Error Rate (EER) of 2.0% across 12,000 ECG beats.
• The system demonstrated effective authentication performance on both within-session and across-session datasets.
• High specificity was maintained due to the two-phase identification process.

Conclusions:

• The developed 1D CNN-based ECG authentication system is effective and highly specific.
• The two-phase approach enhances recognition efficiency and accuracy.
• The system's simplicity and performance make it suitable for practical, real-world biometric applications.