Exposing splicing forgery based on color temperature estimation

Splicing is one of the most common tampering techniques for image manipulation in many forensic cases. Normally color shift in images due to color temperature of illumination can be seen as intrinsic features relative to imaging process. In splicing forgeries, copied area and pasted target image come from different imaging process, and are likely to have different color shift. In this paper, a novel automated authentication method is presented to expose splicing manipulation and locate manipulated areas by discriminating the inconsistencies of color shift in an image. In order to minimize human interaction on detection of splicing forgeries as well as localization of manipulated areas, a forensic image is divided into blocks with grid-based strategy. After calculation on color temperature of each blocks with white-point algorithm, reference color temperature is obtained with a random restricted algorithm. Then color temperature distance between each block and reference area is calculated sequentially. At last, by comparing color temperature distance with an optimized threshold determined by OSTU algorithm. This method enables us to judge if splicing has occurred and furthermore localize manipulated area simultaneously. Experiments show that the proposed method can speed up the quantitative detection of possible splicing manipulation and localize manipulated area automatically.