Painter classification is a well-known research field in academic AI. The project presented here not only predicts artists with an accuracy of more than 90 percent but also levels out huge class imbalances. The main is to detect fake or counterfeit images. Two approaches are pursued. Firstly, 160 images from the “century-counterfeiter” Wolfgang Beltracchi were entered into the dataset. About a third of these images originate from Beltracchi’s criminal years. They were made available by the LKA Berlin for research purposes. The model predicts Beltracchi images with an accuracy comparable to other painters and thus suggests that each artist does indeed have a digital fingerprint, regardless of whether he or she imitates other artists. The model has also been altered to recognize fake images of which the counterfeiter is unknown, or where the author is controversial, like with Leonardo da Vincis „Salvator Mundi“, the most expensive painting ever auctioned. Curious? Come to the presentation to find out more!
Wolfgang Reuter (Alexander Thamm GmbH)