Initial Study Using Electrocardiogram for Authentication and Identification.

Area of Science:

• Biometrics
• Physiological computing
• Signal processing

Background:

• Electrocardiogram (ECG) is increasingly recognized for its potential as a physiological signature in biometric systems (BS).
• Previous research highlights ECG's unique characteristics for individual identification and authentication.
• Standardization and comparison of different ECG-based biometric approaches are needed.

Purpose of the Study:

• To investigate the efficacy of ECG as a biometric trait for individual identification and authentication.
• To compare different signal templates (cardiac cycles and scalograms) and machine learning methodologies for optimizing ECG-based biometrics.
• To evaluate the performance of ECG biometrics using a public dataset with longitudinal data.

Main Methods:

• Utilized the CYBHi public database with two off-the-person ECG records from 63 subjects.
• Generated two types of biometric templates: cardiac cycles (CC) and scalograms.
• Employed Linear Discriminant Analysis (LDA), k-Nearest Neighbors (kNN), Decision Trees (DT), Support Vector Machines (SVM), and Convolutional Neural Networks (CNN) for identification.
• Implemented a distance-based algorithm with leave-one-out cross-validation for authentication and impostor evaluation.

Main Results:

• Identification accuracies reached 79.37% using CC with LDA and 69.84% using scalograms with CNN.
• Authentication accuracy was 90.48% for CC and 98.42% for scalograms.
• Impostor scores were 13.06% for CC and 14.34% for scalograms, indicating robust performance.

Conclusions:

• ECG signals can be effectively utilized for reliable personal recognition in biometric systems.
• The comparison of templates and methodologies provides valuable insights for optimizing ECG-based biometric system performance.
• This study represents a comprehensive evaluation of ECG biometrics, comparing various techniques for the first time.