Learning an Artist's Style: Just What Does a Pigeon See in a Picasso?

Abstract

Judgments of style in art, music, and literature are commonplace, although the mechanisms providing for this structural sensitivity are not well understood. Watanabe, Sakamoto, and Wakita (1995) showed that pigeons trained to discriminate colour slides of paintings of Picasso from those of Monet could generalise this discrimination not only to new paintings of Picasso and Monet, but also to paintings of other cubist and impressionist painters. These results suggest that the bases for such judgments of artistic style may be simpler than normally thought. This tacit sensitivity to artistic style is explored in terms of a simple PCA network model applied to pixel-maps of the paintings. The eigenvectors obtained from the singular value decomposition of sets of these pixel-maps provide for descriptions of the stimuli in terms of visual “macro-features”. These macro-features provide a simple basis not only for recognising previously-experienced paintings, but for the successful discrimination of novel paintings into various style categories. A summary of simulations of the performance of Watanabe et al.’s pigeons using precisely the same stimuli and tasks is provided. The results suggest that the eigen-decomposition is a necessary first-step, and that the bases for judgments of style may indeed be quite simple.