Painting Authorship and Forgery Detection Challenges with AI Image Generation Algorithms: Rembrandt and 17th Century Dutch Painters as a Case Study

Abstract

Image authorship attribution presents many challenges and difficulties which have increased with the capabilities presented by synthetic image generation through different artificial intelligence algorithms available today. The hypothesis in this research considers the possibility of using artificial intelligence as a tool to detect forgeries through the usage of a deep learning algorithm. The proposed algorithm was trained using a dataset comprised of paintings by Rembrandt and other 17th century Dutch painters. Three experiments were performed with the proposed algorithm. The first was to build a classifier able to ascertain whether a painting belongs to the Rembrandt or non-Rembrandt category, depending on whether it was painted by this author or not. The second tests included other 17th century painters in four categories. Artworks could be classified as Rembrandt, Eeckhout, Leveck or other Dutch painters. The third experiment used paintings generated by Dall-e 2 and attempted to classify them using the prior categories. Experiments confirmed the hypothesis with best executions reaching accuracy rates of more than 90%. Future research with extended datasets and improved image resolution are suggested to improve the obtained results.