Deepfake Image Detection using Deep Learning

Deepfake images are generated by modifying existing visuals and are frequently exploited in harmful ways. When executed proficiently, these images can be almost indistinguishable from authentic ones. The increasing development of deep learning techniques has largely fueled the increase in deepfake content. While numerous techniques are available for creating deepfake images, the most commonly employed are GANs and autoencoders. This paper presents a way to make deepfake detection more accurate by using a combination of the YOLOv8 model and a Recurrent Neural Network (RNN). YOLOv8 extracts spatial details from the images, while the RNN identifies temporal patterns and subtle irregularities across image frames, signaling potential manipulation. This methodology presents an efficient solution for identifying deepfake images and reducing their misuse.