Inferring compositional style in the neo-plastic paintings of Piet Mondrian by machine learning

Abstract

We trained generative models and decision tree classifiers with positive and negative examples of the neo-plastic works of Piet Mondrian to infer his compositional principles, to generate "faux" works, and to explore the possibility of computer-based aids in authentication and attribution studies. Unlike previous computer work on this and other artists, we used "earlier state" works-intermediate versions of works created by Mondrian revealed through x-radiography and infra-red reflectography-when training our classifiers. Such intermediate state works provide a great deal of information to a classifier as they differ only slightly from the final works. We used methods from machine learning such as leave-one-out cross validation. Our decision tree classifier had accuracy of roughly 70% in recognizing the genuine works of Mondrian versus computer-generated replicas with similar statistical properties. Our trained classifier reveals implicit compositional principles underlying Mondrian's works, for instance the relative visual "weights" of the four colors (red, yellow, blue and black) he used in his rectangles. We used our trained generative model to generate "faux" Mondrians, which informally possess some of the compositional attributes of genuine works by this artist.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 7 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.