Modular Convolutional Neural Network for Discriminating between Computer-Generated Images and Photographic Images

Huy Nguyen (1) , T. Ngoc-Dung T Tieu (1) , Hoang-Quoc Nguyen-Son (1) , Vincent Nozick (2) , Junichi Yamagishi (1) , Isao Echizen (1)

1 NII - National Institute of Informatics
2 JFLI - Japanese French Laboratory for Informatics

Discriminating between computer-generated images (CGIs) and photographic images (PIs) is not a new problem in digital image forensics. However, with advances in rendering techniques supported by strong hardware and in genera-tive adversarial networks, CGIs are becoming indistinguishable from PIs in both human and computer perception. This means that malicious actors can use CGIs for spoofing facial authentication systems, impersonating other people, and creating fake news to be spread on social networks. The methods developed for discriminating between CGIs and PIs quickly become outdated and must be regularly enhanced to be able to reduce these attack surfaces. Leveraging recent advances in deep convolutional networks, we have built a modular CGI-PI discriminator with a customized VGG-19 network as the feature extractor, statistical convolutional neural networks as the feature transformers, and a discriminator. We also devised a probabilistic patch aggregation strategy to deal with high-resolution images. This proposed method outper-formed a state-of-the-art method and achieved accuracy up to 100%.

Mots clés: convolutional neural network; digital image forensics; computer-generated image; photographic image

Domaines: Vision par ordinateur et reconnaissance de formes [cs.CV]; Traitement des images [eess.IV]

Huy Nguyen, T. Ngoc-Dung T Tieu, Hoang-Quoc Nguyen-Son, Vincent Nozick, Junichi Yamagishi, et al.. Modular Convolutional Neural Network for Discriminating between Computer-Generated Images and Photographic Images. 13th International Conference on Availability, Reliability and Security, Aug 2019, Hambourg, Germany. pp.1-10, ⟨10.1145/3230833.3230863⟩. ⟨hal-02007485⟩