Image splicing detection based on general perspective constraints

Abstract

Image Forensics offers numerous solutions for authenticating the contents of digital images. Unfortunately most of these technologies are ready to work only in controlled environments and their performance heavily drop when applied in real world scenario (e.g, social network, low resolution images, ...) where the images have gone through a chain of unknown processes. In this paper we present a method for forgery detection based on perspective constraints; similar techniques have been proposed in the past but they are effective only when the image is captured with no tilt and no roll thus been unusable in most natural scenes. Here, this solution is extended to include these cases, and we show its applicability even when the image is exchanged through a social network (specifically Facebook and Twitter) where the image is subjected to heavy compression and resizing.