LAUSR

With the rapid development of Computer Graphics, the computer-generated images (CG) are almost as realistic as real photographs(PG) and it is difficult to distinguish between CG and PG accurately with the naked eye. Image is an important carrier for people to get information on a daily basis. However the spread of CG produced for malicious purposes may disrupt social order and even undermine social stability. Therefore, the accurate detection of CG and PG is of great significance. In this paper, we (1) introduce 11 approaches that apply deep learning to the implementations of CG detection, and divide them into 4 categories based on the network structure; (2) give an introduction to the available datasets; (3) design a series of experiments to test the detection performance of each approach,then analyze the experimental results; The experimental results show that most approaches can differentiate CG from PG, while the detection accuracy and efficiency of each model are different. Nevertheless none of these methods is valid when the images tampered by noise. Above all (4) summarize the problems and challenges in this field, and look forward to the trends in future research.