What makes the difference in visual styles of comics: From classification to style transfer

Abstract

The recent success of deep neural network in computer vision provided a new framework to detect visual features of painting styles. However, most deep learning-based approaches analyzing artworks are not interested in popular arts such as comics. In this works, we investigate the artistic styles of comics with deep neural networks. First, we classify comic book pages into five different artists using Convolutional Neural Networks. And the internal features of comic styles are then captured via a feature visualization technique. Second, a style transfer algorithm is applied to several comic book pages drawn by three different artists. We verify how the visual property of a style is transferred to a page using several examples. This is one of the first attempts to analyze in detail the styles of comics with deep neural networks.