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

Electrocardiograms (ECGs) have emerged as promising biometric traits, offering enhanced security through their intrinsic and dynamic properties. These characteristics make ECGs more resistant to theft or forgery compared to traditional biometrics. However, like other biometric systems, ECG biometrics remain vulnerable to hostile attacks that often start with the use of a complete ECG trace. One straightforward method to generate such a trace involves concatenating reconstructed heartbeats to form continuous waveforms. This approach, however, is easily detected by basic presentation attack detection techniques, leading to an unrealistic evaluation of the biometric system's security and susceptibility to attacks. To address this limitation, this study proposes a conditional generative adversarial network-based approach to generate synthetic continuous ECG traces with dynamic variability, including inter-beat and heart rate changes. By employing adversarial training, the generator produces realistic ECG traces capable of compromising a principal component analysis-based ECG biometric system. Experimental results reveal that this approach increases the false-positive identification error rate by 5.75%, highlighting both the effectiveness of the proposed method and the vulnerability of the ECG biometric system to non-zero effort attacks.