2017/11/28 by Edmar Rezende, Guilherme Ruppert, de Rezende, Edmar R. S. +5
Biochemistry, Genetics and Molecular Biology · Computer Science · #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #Digital Media Forensic Detection #FOS: Computer and information sciences #Generative Adversarial Networks and Image Synthesis
paper · pdf · doi:10.48550/arxiv.1711.10394
openalex publication_date 2017/11/28 · openalex created_date 2022/09/28 · openalex updated_date 2026/07/28
The recent computer graphics developments have upraised the quality of the\ngenerated digital content, astonishing the most skeptical viewer. Games and\nmovies have taken advantage of this fact but, at the same time, these advances\nhave brought serious negative impacts like the ones yielded by fakeimages\nproduced with malicious intents. Digital artists can compose artificial images\ncapable of deceiving the great majority of people, turning this into a very\ndangerous weapon in a timespan currently know as Fake News/Post-Truth" Era. In\nthis work, we propose a new approach for dealing with the problem of detecting\ncomputer generated images, through the application of deep convolutional\nnetworks and transfer learning techniques. We start from Residual Networks and\ndevelop different models adapted to the binary problem of identifying if an\nimage was or not computer generated. Differently from the current\nstate-of-the-art approaches, we don't rely on hand-crafted features, but\nprovide to the model the raw pixel information, achieving the same 0.97 of\nstate-of-the-art methods with two main advantages: our methods show more stable\nresults (depicted by lower variance) and eliminate the laborious and manual\nstep of specialized features extraction and selection.\n