Comparative study of DL-based methods performance for camera model identification with multiple databases

Camera identification is an important topic in the field of digital image forensics. There are three levels of classification: brand, model, and device. Studies in the literature are mainly focused on camera model identification. These studies are increasingly based on deep learning (DL). The methods based on deep learning are dedicated to three main goals: basic (only model) - triple (brand, model and device) - open-set (known and unknown cameras) classifications. Unlike other areas of image processing such as face recognition, most of these methods are only evaluated on a single database (Dresden) while a few others are publicly available. The available databases have a diversity in terms of camera content and distribution that is unique to each of them and makes the use of a single database questionable. Therefore, we conducted extensive tests with different public databases (Dresden, SOCRatES, and Forchheim) that combine enough features

to perform a viable comparison of DL-based methods for camera model identification. In addition, the different classifications (basic, triple, open-set) pose a disparity problem preventing comparisons. We therefore decided to focus only on the basic camera

model identification. We also use transfer learning (specifically fine-tuning) to perform our comparative study across databases.