An increasing number of studies are investigating how to automatically recognize painters from digital artwork images. We approach this problem in a supervised manner, by training a high-capacity convolutional neural network, capable of predicting a large number of artists from low-resolution scans. We evaluate the proposed solution in a Kaggle competition, in which pairs of paintings need to be classified based on the identity of their authors. The main contribution of our work is the provision of empirical evidence that themes and motifs, similar to low-level features, contain discriminative potential for identifying painters.
|
