Identifying Individual Camera Device From RAW Images

This paper investigates the problem of identifying the individual imaging device source of the same model from a natural image in RAW format. We propose an enhanced Poissonian-Gaussian model describing the distribution of pixels from a RAW image. The parameters of this statistical noise model are considered as unique fingerprints and, hence, used to identify the source camera device. The source camera identification problem is cast within the framework of the hypothesis testing theory. In an ideal context, where all model parameters are perfectly known, the Likelihood Ratio Test (LRT) is presented, and its performance is theoretically established. The statistical performance of this optimal LRT serves as an upper bound for the detection power. For a practical use, when the nuisance parameters are unknown, a generalized LRT based on estimation of those parameters is designed to deal with unknown expectation of pixels (roughly, the image content). More importantly, by combining multiple sub classifiers together with the voting strategy, a multi-classifier is proposed to identify multiple individual devices. Numerical results on simulated data and on various large dataset of real natural images highlight the relevance of our proposed approach.