Stylometry of paintings using hidden Markov modelling of contourlet transforms

Visual stylometry is the task of quantifying artistic style in the visual arts. In this paper we present a method for visual stylometry of paintings from digital reproductions. Our method is framed around modelling contourlet transforms of the digital reproductions with hidden Markov models. Using the contourlet transform in the field of classification is a new approach motivated by the contourlets’ efficiency in representing piecewise smooth contours such as brushstrokes. To test our method we have used paintings related to the Danish painter Asger Jorn and drawings related to the Flemish artist Pieter Bruegel the Elder. The paintings related to Asger Jorn are recorded in multiple digital images and by two different cameras. With multiple sources we are able to get insight into the robustness of our method against different means of acquisition. Through a cross-validation of the Jorn images by one of the cameras we are able to correctly classify 39 out of 44 images; based on this classifier we can correctly classify 28 out of 36 images in the other data set. A cross-validation of the Bruegel images correctly classifies 11 out of 13 images.