Do GANs Leave Artificial Fingerprints?

By Francesco Marra, Diego Gragnaniello, Luisa Verdoliva, Giovanni Poggi. 2018

397 citations recorded by the source corpus.

In the last few years, generative adversarial networks (GAN) have shown tremendous potential for a number of applications in computer vision and related fields. With the current pace of progress, it is a sure bet they will soon be able to generate high-quality images and videos, virtually indistinguishable from real ones. Unfortunately, realistic GAN-generated images pose serious threats to security, to begin with a possible flood of fake multimedia, and multimedia forensic countermeasures are in urgent need. In this work, we show that each GAN leaves its specific fingerprint in the images it generates, just like real-world cameras mark acquired images with traces of their photo-response non-uniformity pattern. Source identification experiments with several popular GANs show such fingerprints to represent a precious asset for forensic analyses.

Subfields: Computer Vision and Pattern Recognition, Generative Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

CNN-Generated Images Are Surprisingly Easy to Spot… for Now (2019, methodology): “Following prior work, we perform a simple form of high-pass filtering (subtracting the image from its median blurred version) before calculating the Fourier transform, as it provides a more informative visualization [25].”
DIRE for Diffusion-Generated Image Detection (2023, neutral): “[27] and Yu et al .”
Detecting CNN-Generated Facial Images in Real-World Scenarios (2020, background): “As shown by [2, 29, 52], CNN-generated images have pixel patterns dissimilar to real images, which might become more distinctive by learning more intrinsic (pixel-level) image features, such that detection models might generalize better to unseen (e.”
Artificial Fingerprinting for Generative Models: Rooting Deepfake Attribution in Training Data (2020, background): “[34] shows that generative models leave unique noise residuals to generated samples, which allows deepfake detection.”
GANprintR: Improved Fakes and Evaluation of the State of the Art in Face Manipulation Detection (2019, background): “Considering that the ﬁngerprints are high frequency signals [31], we hypothesise that their removal could be performed by an autoencoder, which acts as a non-linear low-pass ﬁlter.”
Source Generator Attribution via Inversion (2019, methodology): “[19] have recently presented methods which use noise-type signatures to attribute a given image to a specific generator, which is conceptually similar to how Photo Response NonUniformity (PRNU) [16] provides source camera identification.”
Attributing Fake Images to GANs: Learning and Analyzing GAN Fingerprints (2018, methodology): “Comparisons also show that, in a variety of conditions, our learned fingerprints are consistently superior to the very recent baseline [45] for attribution, and consistently outperform inception features [52] for cross-source distinguishability.”
Media Forensics and DeepFakes: An Overview (2020, methodology): “using available software [170], [180] is available at [228].”
Learning on Gradients: Generalized Artifacts Representation for GAN-Generated Images Detection (2023, background): “[31] believe that every GAN model has unique fingerprints after training, and extracting it from generated images to perform detection.”
Leveraging Frequency Analysis for Deep Fake Image Recognition (2020, methodology): “…in their original representation (i. e., raw pixels) and after applying DCT: K-nearest-neighbor, Eigenfaces (Sirovich & Kirby, 1987)), a CNN-based classiﬁer developed by Yu et al. (2019a), and a steganalysis method based on photo-response non-uniformity (PRNU) patterns by Marra et al. (2019).”
Face X-Ray for More General Face Forgery Detection (2019, background): “, lossy compression schemes that introduce consistent blocking artifacts, GAN based synthesis algorithms that may leave unique imprints [26, 50].”
Noiseprint: A CNN-Based Camera Model Fingerprint (2018, background): “For example, completely synthetic images can be characterized by artificial fingerprints [76] and the use of a siamese network procedure can be helpful to highlight them.”
Incremental learning for the detection and classification of GAN-generated images (2019, background): “These artificial fingerprints have been studied in [11], [12], they have been shown to characterize not only a given architecture but also a specific instantiation of it, e.”
On The Detection of Synthetic Images Generated by Diffusion Models (2022, methodology): “Previous work [7, 8] established the existence of GAN fingerprints and their dependence on both the GAN architecture (number and type of layers) and its specific parameters (filter weights).”
The Creation and Detection of Deepfakes (2020, background): “03 [125] 2019 • • • Statistics • • Statistics & Steganalysis [128] 2019 • • • • PRNU * • 90.”
Global Texture Enhancement for Fake Face Detection in the Wild (2020, background): “Recently, some researchers have investigated the problem of fake face detection [17, 26, 27, 23, 24, 32, 34, 30].”
Detecting GAN generated Fake Images using Co-occurrence Matrices (2019, background): “The detection of GAN images is a new area in image forensics and there are very few papers in this area [36, 37, 38, 39, 40, 41, 42, 43, 44, 45].”