A GAN-based Approach to Detect AI-Generated Images

The ease with which deep learning can generate fake images has created a pressing need for a robust platform to distinguish between real and fake imagery. However, existing methods in image forensics rely on complex deep learning architectures that are expensive to train and have limited usability due to their large model size. This study examines the difficulty of detecting state-of-the-art image manipulations, both manually and automatically. We introduce G-JOB GAN, a machine learning model based on Generative Adversarial Networks (GAN), which generates highly realistic images and achieves a 95.7% accuracy in detecting realistic generated images. The same architecture of G-JOB Gan can also detect fake images with a similar probability. To verify the results, we compare our results to several similar GAN architectures, including Style GAN, Pro GAN, and the Original GAN. Our model outperforms other GAN models in term of detection accuracy.