Analysis of artistic styles in oil painting using deep-learning features

The styles of oil paintings are pre-defined from an aesthetic perspective along with the historical background. With features learnt by Convolutional Neural Networks (CNNs), we construct two style spaces to visualize the distributions of styles from both low-level and high-level features. Besides, the distance maps are generated to illustrate the relationship between chosen styles. We validate that the relations between artistic styles in machine learning correspond with those in genuine art history. Our distance metrics are 83.3% accurate based on the subjective assessment of the art experts.