Description
The forensic detection of resampling operations in 1D and 2D signals is of great interest in multimedia forensics, as it can indicate that the object under investigation has been resized or subject to other geometric transformations (e.g., rotation in case of images). Due to the limited effect introduced in the signal, the detection of light resampling (i.e, when the image is just slightly shrinked or enlarged) represents a challenging task for existing techniques. For the case of light downscaling, recent theoretical advances have quantified the amount of forensic information available, thus giving an indication on the achievable detection performance under certain assumptions.
The goal of this thesis is to relate the information-theoretical results to the detection performance of state-of-the-art detectors on synthetic and real signals, where the relationship between different parameters will be taken into account.