PRNU-Based Source Camera Statistical Certification

The problem of detecting the presence of a PRNU pattern in a query image can be stated as a hypothesis testing problem. The test statistics that have been proposed to perform this test all suffer from the same drawback: decision thresholds need to be set empirically. This poses a major problem for source camera certification, since these methods do not provide an accurate false alarm rate for each detection but rather a lower bound related to the size of the dataset used to derive such thresholds. In this work, we propose an alternative approach for source camera certification that can be used together with the classic testing strategies. Our method relies on two hypothesis tests based on local correlations which do not require computing empirical distributions. The p-value of the test serves is a statistically founded confidence measure that can serve as certification. Our results show that in most cases, the PRNU true detections give almost absolute guarantees, with p-values smaller than 10^−100, while most true negatives deliver p-values above 10^−1.