Digital Image Forgery Detection by Local Statistical Models

We propose an application of local statistical models in the form of a locally estimated Gaussian mixture to image forgery detection. The estimated mixture is used to compute the so called log-likelihood transformation of the original image. We show that image manipulations of different type may be visible in a suitably designed log-likelihood image. Unlike other methods the forgery detection based on local statistical model is rather non-specific and suitable to emphasize various traces of possible image tampering without any prior information.