Deep Learning Model for Detecting Copy-Move Attack in Images: Testing and Verification

The modern level of digital technology leads to the widespread use of multimedia content: photo and video images, audio, texts, presentations, etc. This trend is significantly enhanced with the development of Internet technologies. Today, every gadget connected to the Internet continuously creates, processes, and sends huge amounts of multimedia information: in social networks; news channels; advertising mailings; messengers and much more. Obviously, this greatly simplifies communication between people and provides convenient, fast, and reliable information services. However, every new technology can be used for damaging purposes. For example, the Internet is currently overflowing with fake photo and video content. Fake multimedia has an extremely negative aspect: it devalues intellectual property and copyright; it compromises objective journalism; it causes reputational and material damage and much more. All this makes to develop and constantly improve new technologies for detecting fakes, verify them repeatedly and test them in various Internet applications. This article explores a well-known type of image spoofing based on copy-move attack. This simple attack can be implemented quickly even in automatic mode, but it is extremely difficult to detect fake images in a huge stream of multimedia data. We consider a deep learning model using convolutional neural networks and perform numerous tests on different datasets. We show that this approach can indeed be used to detect some fake images. However, to build a universal protection mechanism, it is necessary to significantly extend the datasets and take into account the peculiarities of copy-move attacks.