Incremental learning for the detection and classification of GAN-generated images

Marra, Francesco; Saltori, Cristiano; Boato, Giulia; Verdoliva, Luisa

doi:10.1109/wifs47025.2019.9035099

Cited by 99 publications

(55 citation statements)

References 24 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the algorithm is time-consuming, hard to be deployed in real systems, and the performance is still far from satisfactory. Marra et al [25] detected fake images with incremental learning. However, it only works when many GAN models are accessible in the training phase.…”

Section: Related Workmentioning

confidence: 99%

Global Texture Enhancement for Fake Face Detection in the Wild

Liu

Torr

2020

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

164

View full text Add to dashboard Cite

Generative Adversarial Networks (GANs) can generate realistic fake face images that can easily fool human beings. On the contrary, a common Convolutional Neural Network (CNN) discriminator can achieve more than 99.9% accuracy in discerning fake/real images. In this paper, we conduct an empirical study on fake/real faces, and have two important observations: firstly, the texture of fake faces is substantially different from real ones; secondly, global texture statistics are more robust to image editing and transferable to fake faces from different GANs and datasets. Motivated by the above observations, we propose a new architecture coined as Gram-Net, which leverages global image texture representations for robust fake image detection. Experimental results on several datasets demonstrate that our Gram-Net outperforms existing approaches. Especially, our Gram-Net is more robust to image editings, e.g. down-sampling, JPEG compression, blur, and noise. More importantly, our Gram-Net generalizes significantly better in detecting fake faces from GAN models not seen in the training phase and can perform decently in detecting fake natural images.

show abstract

Section: Related Workmentioning

confidence: 99%

Global Texture Enhancement for Fake Face Detection in the Wild

Liu

Torr

2020

2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)

164

View full text Add to dashboard Cite

show abstract

“…More recent works such as [17,18,19] extend the analysis to the frequency domain, where the upsampling step in the GAN generation leaves specific artifacts. The second category of GAN synthesized face detection methods are of data-driven nature [20,21,22,23,24], where a deep neural network model is trained and employed to classify real and GAN-synthesized faces. Methods of the third category look for physical/physiological inconsistencies by GAN models.…”

Section: Related Workmentioning

confidence: 99%

Exposing GAN-Generated Faces Using Inconsistent Corneal Specular Highlights

Lyu

2021

ICASSP 2021 - 2021 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)

View full text Add to dashboard Cite

Sophisticated generative adversary network (GAN) models are now able to synthesize highly realistic human faces that are difficult to discern from real ones visually. In this work, we show that GAN synthesized faces can be exposed with the inconsistent corneal specular highlights between two eyes. The inconsistency is caused by the lack of physical/physiological constraints in the GAN models. We show that such artifacts exist widely in high-quality GAN synthesized faces and further describe an automatic method to extract and compare corneal specular highlights from two eyes. Qualitative and quantitative evaluations of our method suggest its simplicity and effectiveness in distinguishing GAN synthesized faces.

show abstract

“…As the imbalance in the current dataset stands, (10) also implements a similar approach where a GAN is trained to output artificial minority class data which are combined with the original training dataset. This forms the augmented training set that is used to improve classifier performance.…”

Section: Related Workmentioning

confidence: 99%

Optimized Stacking Ensemble (OSE) for Credit Card Fraud Detection using Synthetic Minority Oversampling Model

Veigas¹,

Regulagadda²,

Kokatnoor³

2021

IJST

View full text Add to dashboard Cite

Objectives: Credit fraud is a global threat to financial institutions due to specific challenges like imbalanced datasets and hidden patterns in real-life scenarios. The objective of this study is to propose a model that effectively identifies fraudulent transactions. Methods: Methods such as Synthetic Minority Oversampling Technique (SMOTE) and Generative Adversarial Networks (GAN) that artificially generate synthetic data are used in this paper to approximate the distribution of data among the two classes in the original dataset. After balancing the dataset, the individual models Multi-Layer Perceptron (MLP), k-Nearest Neighbors algorithm (kNN) and Support Vector Machine (SVM) are trained on the augmented dataset to establish an initial improvement at the data level. These base-classifiers are further incorporated into the Optimized Stacked Ensemble (OSE) learning process to fit the meta-classifier which creates an effective predictive model for fraud detection. All base-classifiers and the final Optimized Stacked Ensemble (OSE) have been implemented to critically assess and evaluate their performances. Findings: Empirical results obtained in this paper show that the quality of the final dataset is considerably improved when Synthetic Minority Oversampling Technique (SMOTE) and Generative Adversarial Networks (GAN) are used as oversampling algorithms. The Multi-Layer Perceptron model showed an increase of 10% in the F1 Score while kNN and SVM showed an increase of 3% each. The optimized model is built using a Stacking Classifier that combines the GAN-improved Multi-Perceptron Model with the other standard classification models such as KNN and SVM. This ensemble outperforms the existing enhanced Multi-Layer Perceptron with near-perfect accuracy (99.86%) and an increase of 16% in F1 Score, resulting in an effective fraud detection mechanism. Novelty: For the current dataset, the Optimized Stacked Ensemble model shows an increase of 16% in F1 Score as compared to the existing Multi-Perceptron model.

show abstract

Incremental learning for the detection and classification of GAN-generated images

Cited by 99 publications

References 24 publications

Global Texture Enhancement for Fake Face Detection in the Wild

Global Texture Enhancement for Fake Face Detection in the Wild

Exposing GAN-Generated Faces Using Inconsistent Corneal Specular Highlights

Optimized Stacking Ensemble (OSE) for Credit Card Fraud Detection using Synthetic Minority Oversampling Model

Contact Info

Product

Resources

About