ML Security PapersWavelet-based GAN Fingerprint Detection using ResNet50
Sai Teja Erukude , Suhasnadh Reddy Veluru , Viswa Chaitanya Marella
Published on arXiv
2510.21822
Output Integrity Attack
OWASP ML Top 10 — ML09
Key Finding
Daubechies wavelet preprocessing with ResNet50 achieves 95.1% accuracy on StyleGAN vs. real image classification, a 13.6 percentage point gain over the spatial-domain baseline (81.5%).
DWT-ResNet50 GAN Fingerprint Detector
Novel technique introduced
Identifying images generated by Generative Adversarial Networks (GANs) has become a significant challenge in digital image forensics. This research presents a wavelet-based detection method that uses discrete wavelet transform (DWT) preprocessing and a ResNet50 classification layer to differentiate the StyleGAN-generated images from real ones. Haar and Daubechies wavelet filters are applied to convert the input images into multi-resolution representations, which will then be fed to a ResNet50 network for classification, capitalizing on subtle artifacts left by the generative process. Moreover, the wavelet-based models are compared to an identical ResNet50 model trained on spatial data. The Haar and Daubechies preprocessed models achieved a greater accuracy of 93.8 percent and 95.1 percent, much higher than the model developed in the spatial domain (accuracy rate of 81.5 percent). The Daubechies-based model outperforms Haar, showing that adding layers of descriptive frequency patterns can lead to even greater distinguishing power. These results indicate that the GAN-generated images have unique wavelet-domain artifacts or "fingerprints." The method proposed illustrates the effectiveness of wavelet-domain analysis to detect GAN images and emphasizes the potential of further developing the capabilities of future deepfake detection systems.
Key Contributions
- Wavelet-domain preprocessing pipeline using Haar and Daubechies DWT to expose GAN-specific frequency artifacts before classification
- Empirical demonstration that wavelet-preprocessed ResNet50 (95.1%) substantially outperforms spatial-domain ResNet50 (81.5%) for StyleGAN detection
- Evidence that GAN images carry unique wavelet-domain fingerprints exploitable by standard deep classifiers
🛡️ Threat Analysis
Proposes a novel deepfake/GAN-generated image detection method — a forensic technique for verifying output authenticity and integrity of AI-generated visual content. Detection of AI-generated images (deepfakes) is explicitly covered under ML09 output integrity.