“…Most previous work in camera device identification focuses on developing what we term a universal detector [7,9,11,12,13,14,15] where the authors aim to create a single system that works for any questioned image and camera device. Over the years, universal detectors that use the PRNU have been tested against cropping and scaling [13], gamma correction and denoising [6], compression [14], lens distortion [16], and contrast enhancement, histogram equalization, and white balance [15]. Most previous work determines if a questioned image I u came from a specific known camera C k by comparing the value of interest δ = Δ( X u , I u F k ) from Equation (2) to an ad‐hoc decision threshold t , where t is typically chosen based on a constructed set of similarity scores between an image and a camera fingerprint known to be from the same camera (matching) and an image and a camera fingerprint known to be from different cameras (non‐matching).…”