Deep-BIF: Blind Image Forensics Based on Deep Learning

Baole Wei, Min Yu, Kai Chen, Jianguo Jiang

Abstract: Digital images are widely used in all aspects of human society because of their intuitiveness, such as computer forensics and scientific research. Tampering digital images in special fields maliciously, may change the information contained therein, form false information, and cause harm to society. At present, most of the mainstream image forensics algorithms are limited by their dependence on image capture devices or lack of robustness to complex images. We hope to get rid of the above limitations by virtue of the excellent feature extraction ability of deep learning. In this work, we proposed a novel method for blind image forensics analysis based on convolutional neural networks named Deep-BIF. Furthermore, we integrated the rich models for steganalysis of digital images into our network, for the purpose of guiding network training progress with several artificial prior knowledge. We tested our method on CASIA v2.0 dataset and achieved 0.976 in terms of accuracy.