Automatic JPEG Grid Detection with Controlled False Alarms, and Its Image Forensic Applications

With the progress of image manipulation tools and the proliferation of fake news and images posted online on social networks, automatic identification of fake content is becoming indispensable. Lossy image compression leaves traces which can be used to recover the history of an image and to help decide about its authenticity. We propose a new JPEG grid detection algorithm. This operation is the first step of many forensic, anti-forensic, and deblocking algorithms. Our analysis is based on the detection of the blocking artifacts and is global and local at the same time. It retrieves the origin of the JPEG grid in all image regions and detects suspicious discrepancies. Our work is based on the a-contrario framework which reins in the over-detections caused by multiple testing. It also yields a Number of False Alarms (NFA) which gives extremely secure guarantees for tampering detection. We demonstrate the performance of the proposed method with both quantitative and visual results from well-known image databases.