Abstract:
Deep Learning has shown to be helpful in tackling a range of complex problems that we could not have predicted only a few decades ago. Despite the various benefits provided by technology, there are also ways in which it might be used to damage our society. Deepfakes have been identified as one such issue, now a days creating a fake image is so simple using any free website and free mobile application, some security measures are evaluated to confirm whether the image is fake or real, and to solve that problem challenging the reliability of information on the web. Despite the fact that Deep fakes created by neural networks appear to be as authentic as a legitimate picture, they regrettably leave behind spatial and temporal traces or handprints after moderation, which may be detected using a neural network trained to identify Deep fakes. We tried comparing multiple state-of-the-art neural networks (ResNet-50, VGG-19, and XceptionNet) in this study to develop a solution for multiple situations, such as real-time Deepfake identification for use in various social media channels where classification must be accomplished as quickly and efficiently as possible.
