Distinguishing Between AI Images and Real Images with Hybrid Image Classification Methods

Due to the rapid proliferation of artificial intelligence applications, some vulnerabilities in security and ethical issues emerge. With these applications, data such as text, images, audio and video can be easily produced. In order to ensure stability in issues such as security , ethics and quality, it is necessary to identify the data produced by artificial intelligence applications. For this purpose, this study focuses on the classification of images created with artificial intelligence applications and real images. In the study , a dataset containing images produced by artificial intelligence and real images was used. There are a total of 975 images in the dataset. The features of the images in the dataset were extracted with SqueezeNet, InceptionV3 and VGG19 pre-trained CNN (Convolutional Neural Network) models. Classification of features was made with ANN (Artificial Neural Network), KNN (K Nearest Neighbor) and SVM (Support Vector Machine) machine learning methods. The highest classification success was obtained from the InceptionV3+ANN model. It is anticipated that the proposed models can be used to detect images produced with artificial intelligence applications. However, it has been determined that more data is needed to fully solve this challenging task.