Reliable Demosaicing Detection for Image Forensics
Quentin Bammey, Rafael Grompone von Gioi, Jean‐Michel Morel · 2019
Visually plausible image forgeries are easy to create even without particular knowledge or skills. However, most forgeries unknowingly alter the underlying statistics of an image. In particular, demosaicing artefacts created by the camera are usually destroyed or modified when the image is tampered. Most of the literature focus on detecting where these traces are destroyed, and generally do it in a way that still requires a visual interpretation. We introduce an a contrariomethod which detects global demosaicing parameters, and then checks for regions of the image which are inconsistent with these parameters. Detections are guaranteed in the form of a number of false alarms (NFA), which enables the user to control the false positive rate. Such a guarantee is a useful complement to existing methods, and enables inclusion into fully automatic image authentication processes. The source code and an online demo are provided with the article.