A Comparative Review of AI-Generated vs Real Images and Classification Techniques

Images often play a crucial role in art, data analysis, and communication. Images visually represent the physical world. With the development of AI, differentiating between AI-generated and real images has become very difficult over time. So in this review, we have focused on the techniques for forming images, which focus on real images and the images created by Stable diffusion and other such techniques. This paper aims to identify the differences between AI-generated and real images by understanding how they are generated, processed, and represented. A study on CNN techniques that can be utilized to classify AI and real images has been mentioned. The specific architecture of CNN such as VGG, EfficientNet, and ResNet has been explored to find their effectiveness in the classification. In addition, an overview of available datasets has also been provided. This study provides a review of studies that classify AI or real images. This employs approaches such as StyleGAN, DALL-E, and Stable Diffusion. The difficulties that these generation techniques present to the conventional models which include CNN, DenseNet, ResNet, and VGG are also interrogated. While ResNet-50 and Swin-T models achieve good merit in practical applications, classifying images produced by advanced techniques such as ProGAN models is still a challenge. Our analysis incorporates an evaluation and comparison of classification models and their performance as well as directions for future work including the application of EfficientNet and hybrid models aimed at improving propositions on differentiating real and fake images by their accurate classification.