Speaker: Purnachandra Rao (EE10D017)
With accessibility to advanced image editing tools becoming commonplace, creating fake images is turning out to be an effortless task. Image splicing is one form of tampering in which an original image is altered by copying a portion from another source. There are several practical situations in which it is critical to determine whether an image is spliced or not. Among the many cues that can possibly be used to detect the presence of splicing, we employ motion blur as it is a common occurence in hand-held cameras. Specifically, we address the scenario of a static scene in which the cause of blur is due to camera shake. Our approach is grounded on the fact that motion blur (when present) should be consistent across the entire image. Existing methods for dealing with this problem work only in the presence of uniform space-invariant blur. We advance the state-of-the-art by proposing a new methodology that can objectively infer inconsistencies even under space-variant blurring situations. Our method is evaluated on several examples with different types of camera motion and for typical scenes of interest.