This poster was presented at Computer Graphics & Visual Computing (CGVC24) at City St George's, University of London.
Mpho Bwanali, Hassan Ugail, Zied Mnasri, Elias Jensen
Accurate art authentication is crucial for ensuring correct attribution, cultural preservation, and financial valuation of artworks. Machine learning tools can significantly complement established techniques such as connoisseurship, provenance research, and scientific analysis (e.g., pigment analysis and radiographic imaging). This study explores a more objective and explainable approach to machine learning in art authentication. Specifically, we investigate the use of handcrafted features, such as image entropy, energy, texture patterns, and image similarities, to train a OneClassSVM model based on known authentic images of paintings by a given artist. To demonstrate the versatility of our method, we present results for three selected artists: John Constable, Vincent van Gogh, and Jean-Michel Basquiat. We further show by way of an example how our method can be utilised to study a given painting in terms of its attribution to a given artist. These examples, along with the accuracy of our results, indicate the potential of the proposed method as a flexible and efficient tool for art authentication alongside more established methods.