Detecting AI-Generated and Authentic Artworks Using a ResNet50 Convolutional Neural Network Architecture

Abstract

The rapid advancement of generative artificial intelligence (AI) has obscured the distinction between human- and machine-created art, posing significant challenges to authentication, copyright, and artistic integrity. This study addresses the critical need for reliable verification tools by developing and evaluating a deep learning model to automatically classify artworks based on their origin. A ResNet50 Convolutional Neural Network architecture was fine-tuned for the binary classification task. The model was trained on a custom, perfectly balanced dataset comprising 868 images (434 AI-generated, 434 authentic artworks). The training protocol included extensive data augmentation to enhance generalization and an early stopping mechanism to prevent overfitting. The experimental results demonstrate a high level of classification performance. The model achieved a validation accuracy of 86.21%, with a precision of 0.88 and a recall of 0.84 for the AI-Generated class. A Receiver Operating Characteristic (ROC) analysis yielded an Area Under the Curve (AUC) of 0.929, indicating robust discriminative capability. Qualitative error analysis revealed that the model's primary challenges lie in classifying hyper-realistic AI-generated images and authentic artworks with surreal or digitally abstract styles. This study validates the effectiveness of the ResNet50 architecture as a reliable and accessible tool for digital art authentication. It contributes a well-documented performance baseline on a balanced, custom dataset, providing a practical foundation for future research. This work highlights key challenges and suggests future directions, such as the exploration of more advanced architectures and the development of larger, more diverse datasets to further improve detection accuracy.