Photomontage detection with deep learning methods

Today, thanks to the increasing use of social media and the internet almost every photo taken is uploaded to the internet. For this reason, it has become much easier to obtain images. Easily obtaining images has become an encourage to make changes on the image thanks to the developing technologies. In addition to the changes made on the images by the person's own will, the changes made by malicious people cause some problems and can create unwanted situations. With the changes made, it affects a wide area from the violation of personal data to the forgery of any official document. Forgery methods are carried out in different ways. Copy-move method is done by copying an area in the current image to a different area on the same image. Image fusion is created by combining the desired areas of two different images. Image retouching makes the image more interesting by reducing the effect of unwanted areas on the image or by improving the desired area. Based on this idea, in the study, deep learning method was used to determine whether there is a change on the images, and if there is, to mark the area. Thanks to deep learning methods, faster and higher success has been achieved in high-dimensional data. In the study, an application was carried out on the CASIA and CoMoFoD data sets. As a result of the tests performed on the data sets, an average accuracy of 76.86% was obtained.