Unsupervised Clustering for the Chronological Analysis of Digitized Paintings

Abstract

In this study, unsupervised clustering techniques are applied to digitized paintings of two famous Pointillist artists, Paul Signac and Georges Seurat. That way we temp to reveal natural division of each artist oeuvre into different time periods and moreover to unlock the features that mostly contribute to the shaping of clusters and thus to conclude the evolution of the artistic styles adopted by the painter over time. A large set of engineered features is used and the feature ranking process identifies the most important features: run-length features, fractal dimension, and statistical features derived from hue histograms.  Interpretation of the analysis of the most important features shaping the clusters in terms of painters’ style evolution is consistent with evidence of art experts.