Presentation Attacks on the PCA-Based Biometric System Using cGAN-Generated ECGs.

Area of Science:

• Biometrics
• Artificial Intelligence
• Signal Processing

Background:

• Electrocardiograms (ECGs) are promising biometric traits due to their unique intrinsic and dynamic properties, offering better security than traditional methods.
• ECG biometric systems are vulnerable to sophisticated attacks, particularly those using realistic synthetic ECG traces.
• Current methods for generating synthetic ECGs are often detectable, leading to an overestimation of system security.

Purpose of the Study:

• To develop a novel method for generating continuous, synthetic ECG traces with realistic dynamic variability.
• To evaluate the security vulnerabilities of ECG biometric systems against advanced synthetic attack traces.
• To assess the impact of generated synthetic ECGs on the performance of a principal component analysis (PCA)-based ECG biometric system.

Main Methods:

• Utilized a conditional generative adversarial network (cGAN) architecture for synthetic ECG generation.
• Incorporated adversarial training to enhance the realism and dynamic variability (inter-beat and heart rate changes) of generated ECG traces.
• Evaluated the effectiveness of the synthetic ECGs in compromising a PCA-based ECG biometric system.

Main Results:

• The proposed cGAN-based approach successfully generated realistic continuous ECG traces with dynamic variability.
• The synthetic ECG traces were capable of compromising the PCA-based ECG biometric system.
• The false-positive identification error rate increased by 5.75% when attacked with the generated synthetic ECGs.

Conclusions:

• The study demonstrates the vulnerability of ECG biometric systems to advanced synthetic attacks.
• Conditional generative adversarial networks offer a powerful tool for generating realistic synthetic biometric data for security evaluations.
• Further research is needed to develop robust presentation attack detection methods for ECG biometrics.