Attribution and Evaluation of Works of Art: The Application of Artificial Intelligence Techniques to Paintings From the Circle of Rembrandt Van Rijn

Abstract

The present study analyzes the application of Artificial Intelligence techniques to the attribution of authorship of paintings belonging to the circle of Rembrandt van Rijn, a context characterized by strong stylistic similarities. Using the proprietary system Luminari, two case studies were examined: a self-portrait by Rembrandt universally recognized as authentic, and Portrait of a Man (The Auctioneer), a work of controversial attribution. The analysis is based on a dataset of 1,578 images referring to 16 painters of the School of Rembrandt. Different machine learning and deep learning models were compared, including Random Forest, Support Vector Machine, and convolutional neural networks, with particular attention to a ResNet50 pre-trained on ImageNet. The latter showed the best performance, with an overall accuracy of approximately 72%, a significant value in a domain characterized by high authorial ambiguity.

The results highlight both the potential of AI in supporting attribution analysis and the intrinsic limitations in cases of strong stylistic proximity. The introduction of a corrective index related to the School of Rembrandt allows a more cautious and informed evaluation, confirming the authenticity of the self-portrait and suggesting, for The Auctioneer, a non-conclusive attribution within the circle of Aert de Gelder.