AI-Generated Fake Image Detection Using Pre-trained CNN Models

The widespread dissemination of digitally manipulated images, particularly those generated by artificial intelligence (AI), poses a critical challenge to the integrity of visual information in today's digital landscapes. With the increasing accessibility of advanced image-editing technologies, the prevalence of fake images not only inundates digital platforms, but also introduces complexities in discerning genuine content from manipulated countermeasures. This study addresses the urgent need to detect and identify fake images generated by AI by utilizing the CIFAKE dataset. Our approach focused on the application of three pre-trained Convolutional Neural Network (CNN) models: VGG16, MobileNet, and InceptionV4. Through comprehensive testing on the CIFAKE dataset, our findings indicated that the MobileNet model outperformed both VGG16 and InceptionV4 in accurately identifying fake images, achieving an impressive accuracy of 90.10%. This study delves into the effectiveness of pre-trained CNN models, with a specific emphasis on the superior capability of the MobileNet model in recognizing AI-generated fake images. These results provide valuable insights for enhancing strategies to counter the dissemination of deceptive imagery across digital platforms, thereby safeguarding the credibility of visual information.