Image Forensics with Rotation-Tolerant Resampling Detection

Millions of photos are uploaded to social networking sites every day. However, the authenticity of these images has been severely questioned. Recent advances in image editing software have helped forged images spread widely. While these fake images leave no visual clues, they can still be detected by inspecting the traces left by the resampling process. In this paper, we propose a novel rotation-tolerant resampling detection method, and design a blind image forgery detection algorithm based on this resampling detection method. A measurement called "Rate-Distance" is devised for measuring the distance between two resampled images with different resampling history. Images are classified based on their "Rate-Distances". Through experimental results, we demonstrate that the proposed method can achieve high detection accuracy.