Deep learning based forensic applications

Nowadays it has become very easy altering the contents of digital images. These alterations are usually carried out without a bad intention. But in some cases, it is very important to know that a picture is altered or not. Particularly, fake images are created to manipulate political figures. In this context, the trustworthiness of the images is very important in terms of forensic evidence. Current forensic detection methods can produce good results in some cases. However, there are insufficient methods available against most types of alterations. The PRNU-based source device identification method in the forensic detection literature is the most accepted method among similar methods by forensic analysts. In addition, deep learning-based camera model classifier method, which has recently been offered as a solution in this area, proved its success in forensic field. In this study, the deep learning based forensic detection method and the PRNU-based method is examined and a new method based on a special fusion approach is proposed. With this method, the tampered regions on digital images can be detected more accurately than the methods in the literature, even so, the proposed method works well in detecting small-scale forgeries with the size of 100 x 100 pixels.