Deep Fake Image Detection Using Fully Connected Neural Networks

Deep fakes in the social media platform worry the spread of information because it holds a huge scale of misinformation. This paper introduces a CNN-based robust framework capable of being used for detecting Deep Fake images published in social media platforms. It may be possible with innovations in deep learning to use the model to classify whether it is a real image or artificially generated with small anomalies in texture, pixel patterns, and other artifacts that can be impossible for an eye to see. Optimized CNN architecture with performance-wise accuracy objectives; the whole dataset of real and synthesized images will be covered in the training process. This model will include data augmentation techniques as well as multi-layered convolutional filters to be able to handle a tremendous variation of source images, which makes this model adjust well with an image from different quality and format. The experimental results demonstrate that this approach results in extremely high detection accuracy, much beyond what traditional methods could have achieved, indicating great potential for deep fake detection in real-time in a social media scenario. This paper illustrates the design of the model, how to train it, and also analyze its performance to bring forward insights on the implications about countering misinformation and enhancing cybersecurity within cyber domains.