An effective histogram-based approach to JPEG-100 forensics

This paper proposes an improved histogram-based approach to identifying whether an image is never compressed or has undergone JPEG compression with quality factor 100. The key idea is that the image's DCT (Discrete Cosine Transform) coefficients follow either of two families of parametric distributions, corresponding respectively to never compressed images and JPEG-100 compressed ones. This paper highlights that choosing the generalized Gaussian distribution (GGD) to model the DCT coefficients and constructing a DCT histogram with precision higher than integer create a prominent distinction between the DCT coefficients distribution of the two kinds of images. Experiments demonstrate that the proposed approach significantly outperforms existing histogram-based methods for the task of JPEG-100 forensics.