Automatically classifying paintings with perceptual inspired descriptors

We propose a framework for the automatic recognition of artistic genre in digital representations of paintings. As we aim to contribute to a better understanding of art by humans, we extensively mimic low-level and medium-level human perception by relying on perceptually inspired features. While Gabor filter energy has been used for art description, Dominant Color Volume (DCV) and frameworks extracted using anchoring theory are novel in this field. To perform the actual genre recognition, we rely on a late fusion scheme based on combining Multi-Layer Perceptron (MLP) classified data with Support Vector Machines (SVM). The performance is evaluated on an extended database containing more than 4000 paintings from 8 different genres, outperforming the reported state of the art.