Key result
Siamese network using frequency-domain features achieves ~94% accuracy detecting GAN-generated fake ECG signals.
Why the study?
Studies on detecting fake biosignals are still in their early stages, creating a need to analyze fake biosignal characteristics generated by a GAN to detect spoofing attacks in biometric systems.
A Siamese network utilizing frequency-domain features can effectively detect fake ECG signals generated by GANs, achieving 94.41% accuracy to prevent biometric spoofing attacks.
May enable countermeasures to ECG spoofing; leaves open prospective validation in clinical biometric systems.
Fake data can be used for malicious purposes, such as fake news propagation and spoofing attacks. Therefore, studies are underway to detect fake faces, fingerprints, and voice data. However, studies on detecting fake biosignals are still in their early stages. This paper analyzes fake biosignal characteristics generated by a GAN to detect spoofing attacks in biometric systems. To detect checkerboard artifacts in fake biosignal data, the biosignals were analyzed after converting them into frequency bands using a fast fourier transform (FFT) and empirical mode decomposition (EMD). Then, fake biosignals were detected with a Siamese network designed using the analyzed information. According to the experimental results, checkerboard artifacts in the fake electrocardiogram (ECG) signals were predominantly observed in the low-frequency bands. Furthermore, the Cosine Similarity (CS) of IMF (intrinsic mode function)3, calculated from the P, Q, S, and T wave data, was 0.5388. This value is significantly lower than the CS of the raw signal, which was 0.9925. The Siamese network, constructed using frequency-domain features, achieved the highest fake-data detection accuracy of 94.41% when using IMF1–IMF3. This confirmed the effectiveness of the checkerboard artifact information in the frequency bands analyzed in this paper.
No takes yet. Share an insight, caveat, or question.
Kim et al. (2025) studied Spoofing attacks in biometric systems (fake ECG signals). Siamese network with frequency-domain features (FFT and EMD) was evaluated on Fake-data detection accuracy. A Siamese network using frequency-domain features (IMF1–IMF3) achieved 94.41% accuracy in detecting GAN-generated fake electrocardiogram signals.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: