Nineteenth International Conference on the Arts in Society

Abstract

This paper proposes the use of Convolutional Neural Networks (CNNs) as a tool for artwork authenticity verification. The accepted verification process relies on the three-legged stool of provenance, connoisseurship and forensics. We investigate CNNs as a means of authenticating whether a particular artist in fact painted the work attributed to him/her. Our work uses deep learning methodology to study the art of Jean-Michel Basquiat. CNN architecture is used to progressively analyze details of Basquiat’s specific style. We train our tool on high quality image data sets. Once it has been trained, we apply it to the question of whether a particular painting should be attributed to Basquiat. This study uses CNNs for image analysis. We work to create a model capable of accurately distinguishing genuine Basquiat artworks from counterfeit ones. Our model is trained with verified examples of Basquiat’s art. We then apply the deep learning lessons to Basquiat paintings that are in dispute. It is proposed as an additional tool in the authenticity quest. To create the tool we collect quality scans of authenticated Basquiat art. To prepare the images for the model, we resize every artwork to be 256 x 256 pixels. Having all images the same size helps with analysis consistency. We adjust the pixel values of the images to fall between 0 and 1, which helps the training process. To be sure our model can identify a Basquiat, we first test it against other pop and expressionist artworks.