Complementing Connoisseurship With Artificial Intelligence

Abstract

AbstractThough humanistic in spirit and scholarship, the practice of art history has long recognized the value of science‐based investigative techniques, particularly for questions involving artwork authentication. Modes of investigation now include analysis of chemical composition and physical structure in addition to the traditional principles of connoisseurship and provenance research. We propose the use of a new tool − convolutional neural networks − to extend scientific analysis to the visual features of two‐dimensional artwork. Trained on the works of an artist under study and visually comparable works of other artists, our system can identify forgeries and propose attributions. It can also assign classification probabilities within a work, revealing mixed authorship and identifying the contributions of different hands. We describe and illustrate performance in connection with Old Master paintings and drawings. With one interesting exception, our system's attributions match those of authoritative experts for paintings whose authorship has been the subject of longstanding controversy. Although computational, our approach supplements and depends on human expertise in judging an artwork's authenticity and lived history.