Camera model identification Based on hypothesis testing theory

This paper aims to study the problem of imaging device identification using the heteroscedastic property of uncompressed image noise. Noise variance depends on pixels intensity through two parameters which uniquely represent a camera model and hence, enable to identify imaging device. The decision problem is cast in the framework of hypothesis testing theory. First, the theoretical context in which both the inspected image parameters and imaging device properties are known is considered. The most powerful Likelihood Ratio Test (LRT) is presented and its detection performance is analytically calculated. Then, the practical situation when inspected image parameters are unknown, but imaging device properties remain known, is studied. Based on a simple yet efficient image model, the inspected image parameters are estimated. This leads to the designed Generalized Likelihood Ratio Test (GLRT) whose statistical performances are analytically given. Numerical simulations and experimentations on natural images show the relevance of the proposed approach.