Deepfake Detection for Digital Image Security using Deep Learning Methods

Abstract

With the rapid increase and proliferation of digital technologies, manipulated content is also increasing in parallel. With the widespread use of deepfake technology, the detection of manipulated content and deceptive content reduces the risks of manipulated data. This situation leads to serious security consequences at social, political, and personal levels with the creation of fake news, misleading videos, and audio recordings. This technology can also cause serious problems such as malicious use and violation of privacy. Therefore, it is vital to develop preventive measures such as deepfake detection and to use this technology correctly and ethically. The detection of images created using deepfake techniques aims to detect manipulations in media files such as video and audio using artificial intelligence and machine learning techniques. Deepfake detection is usually carried out using deep learning algorithms and models. In this study, a hybrid model consisting of transformer-based networks and Convolutional Neural Networks (CNNs) is used to classify fake and real images. When the results of the study were examined, it was seen that the hybrid model used gave more successful results compared to the literature. The applications were carried out on the Casia-WebFace dataset. According to the results obtained, the proposed artificial intelligence method plays an important role in the classification process of images produced using DeepFake techniques. 98.82% accuracy rate was achieved for the Casia-WebFace dataset. These results show that the proposed artificial intelligence model is effective and successful in predicting deepfake techniques.