On art authentication and the Rijksmuseum challenge: A residual neural network approach

Abstract

The popularity of machine learning algorithms produced numerous applications in the past ten years. One application is that of art authentication which assures that a piece of art is created by an artist. A certificate of authenticity created from proper art authentication significantly increases the value of a piece of art which impacts all parties in an art transaction. The models produced by machine learning algorithms provide an objective measure to authenticate an artist to their artwork collection. In the past ten years numerous machine learning algorithms have been used to address art authentication on a variety of datasets. Our work extends art authentication with residual neural networks and the Rijksmuseum data set. Our results show contributions is four key areas: A performance increase of 21% over the baseline for 958 artists; A new baseline for 1,199 artists; A standard methods for recreating the Rijksmuseum data set; and A standard method for measuring results from imbalanced data for the Rijksmuseum data set.