Exposing image forgery with blind noise estimation

Noise is unwanted in high quality images, but it can aid image tampering. For example, noise can be intentionally added in image to conceal tampered regions and to create special visual effects. It may also be introduced unnoticed during camera imaging process, which makes the noise levels inconsistent in splicing images. In this paper, we propose a method to expose such image forgeries by detecting the noise variance differences between original and tampered parts of an image. The noise variance of local image blocks is estimated using a recently developed technique, where no prior information about the imaging device or original image is required. The tampered region is segmented from the original image by a two-phase coarse-to-fine clustering of image blocks. Our experimental results demonstrate that the proposed method can effectively detect image forgeries with high detection accuracy and low false positive rate both quantitatively and qualitatively.