Image forgeries detection through mosaic analysis: The intermediate values algorithm

Cameras sample each image pixel in one color channel only. The remaining channels are interpolated from neighboring pixels during demosaicing. This operation leaves traces, that can be exploited to authentify images and detect forgeries. This paper describes the method introduced by Choi et al. that exploits the fact that interpolated pixels are more prone to be intermediate values, to detect in which pattern an image has been sampled. We then use this information to find regions that are inconsistent with the global image. We attribute a confidence score to each detection, which can then be thresholded to provide a binary map of detected forgeries.