Detecting Computer Generated Images with Deep Convolutional Neural Networks

摘要: Computer graphics techniques for image generation are living an era where, day after day, the quality of produced content is impressing even more skeptical viewer. Although it a great advance industries like games and movies, can become real problem when application such applied production fake images. In this paper we propose new approach computer generated images detection using deep convolutional neural network model based on ResNet-50 transfer learning concepts. Unlike state-of-the-art approaches, proposed method able to classify between or photo directly from raw data with no need any pre-processing hand-crafted feature extraction whatsoever. Experiments public dataset comprising 9700 show accuracy higher than 94%, which comparable literature reported results, without drawback laborious manual step specialized features selection.