Authentication of splicing manipulation by exposing inconsistency in color shift

Abstract

Splicing is one of the most common tampering techniques for image manipulation in many forensic cases. In this paper, a novel detection method which can expose splicing manipulation by discerning the inconsistencies of color shift in splicing images is presented. Color temperature varies dramatically in formation of different images due to photographic light sources, and always leads to color shift in an image, though it sometimes is imperceptible for human visualization. Therefore color shift can be taken as one kind of optical fingerprint of images for authentication. In this study, we proposed a color shift estimation based method to authenticate the presented images by localizing splicing manipulations automatically. These quantitative evidences contribute a lot to forensic investigators. This method comprises three steps: firstly it estimates the color shift of local partitions and the whole image; secondly exposes inconsistency in color shift by calculating distances between them as an indicator; and finally classifies these partitions into different classes by an optimized threshold. In our proposed method, color shift is estimated by using color constancy algorithm, which has been widely applied in cameras as AWB (automated white balance) to correct color shift in captured images. The following experiments exhibits the effectiveness of the proposed method with visual and quantitative evaluation.