Positional Learning to reverse-engineer the camera processing pipeline and detect image forgeries

As conventional tools and generative-AI-based methods alike can alter images in visually convincing ways, image editing is no longer reserved to experts. However, this ease of manipulation has given rise to malicious manipulation of images, resulting in the creation and dissemination of realistic but fake content to spread disinformation online, wrongfully incriminate someone, or commit fraud. The detection of such forgeries is paramount in exposing those deceitful acts. One approach involves reverse-engineering the image signal processing pipeline, to detect and localize inconsistencies. In this context, positional learning has emerged as a promising and explainable approach to reveal underlying traces of the signal processing pipeline. We show how it can be used to detect forgeries from inconsistencies in the image mosaic or compression history.