Detection of Real Versus Fake Images on Social Media Through Generative Adversarial Networks

One of the most prevalent issues rapidly getting worse on social media is the spread of bogus photos and videos. Social networks’ capillary distribution has made it incredibly simple to disseminate information, including false information. Multimedia content is frequently used to bolster fake news in an effort to increase its legitimacy. Even inexperienced users can effortlessly alter an image with robust image editing programs like Photoshop or GIMP to produce realistic outcomes that elude the notice of human observers. Anyone can create phony photos by adding, removing, or cloning people and objects using commercial media editing programs. Although numerous methods have been put out to identify these types of traditional fakes, new threats are appearing every day. Based on generative adversarial networks (GANs), image-to-image translation seems to be among the riskiest since it allows for highly realistic manipulation of the context and semantics of images. In the present work, we examine the efficacy of various image falsification detectors against image-to-image translation, both under perfect circumstances and with the compression that is sometimes applied during social network uploads. In this study, we examine how well many learning-based techniques perform regarding image-to-image translation identification. Here we examined generative AI approaches, derived from both cutting-edge techniques found in the literature on picture forensics and broadly applicable that have been appropriately trained for this purpose.