Detect digital image splicing with visual cues

Abstract

Image splicing detection has been considered as one of the most challenging problems in passive image authentication. In this paper, we propose an automatic detection framework to identify a spliced image. Distinguishing from existing methods, the proposed system is based on a human visual system (HVS) model in which visual saliency and fixation are used to guide the feature extraction mechanism. An interesting and important insight of this work is that there is a high correlation between the splicing borders and the first few fixation points predicted by a visual attention model using edge sharpness as visual cues. We exploit this idea to develope a digital image splicing detection system with high performance. We present experimental results which show that the proposed system outperforms the prior arts. An additional advantage offered by the proposed system is that it provides a convenient way of localizing the splicing boundaries. © 2009 Springer Berlin Heidelberg.