Data Science in Art: Discerning the Painter’s Hand
Ken Singer, Ambrose Swasey Professor of Physics
Michael Hinczewski, Warren E. Rupp Associate Professor of Physics
Ina Martin, Senior Research Associate (Physics), Adjunct Faculty in the Department of Materials Science and Engineering
Betsy Bolman, Elsie B. Smith Professor in the Liberal Arts and Chair, Department of Art History and Art
The Departments of Art History and Art, Physics, Materials Science and Engineering, the Cleveland Museum of Art and the Cleveland Institute of Art have been collaborating to investigate the application of machine learning (ML) to artist attribution based on confocal optical profilometry data from student-produced painting via the brushstroke texture. A convolutional neural network was applied to classify the surface topography among several students’ paintings. By specifying the painted subject and materials for the students, we were able to carry out a controlled study of various ML approaches including the efficacy of transfer learning as well as scaling, normalization, empirical mode decomposition and other pre-training analyses. We were able to confidently attribute paintings among multiple hands and have been able to determine aspects that inform the learning. Our results suggest potentially significant implications for the art historical field of connoisseurship. To this end, we are now collaborating with the internationally-renown Factum Arte, in a project to apply ML on surface topography in order to test the ability of our techniques to distinguish among the hands of El Greco, his son Jorge, and members of his workshop.