3D lighting-based image forgery detection using shape-from-shading

This paper concentrates on lighting-based forensics. We first show how to fool the forgery detector based on 2D lighting coefficients using a simple counter-forensic strategy. This intermediary result advocates the use of more involved 3D lighting coefficients for forensics purposes. Such a research line means that we need at least an approximation of the 3D surface of the suspect object. Contrary to previous approaches that concentrated on particular kind of shapes (e.g. human faces), we propose a promising approach based on shape-from-shading. This new 3D lighting-based forensic method is more general as the 3D shape is learned from the picture itself. Furthermore, the results are in par with the less general state-of-the-art methods.