Manga Filling Style Conversion with Screentone Variational Autoencoder

摘要

Western color comics and Japanese-style screened manga are two popularcomic styles. They mainly differ in the style of region-filling. However, theconversion between the two region-filling styles is very challenging, andmanually done currently. In this paper, we identify that the major obstaclein the conversion between the two filling styles stems from the differencebetween the fundamental properties of screened region-filling and coloredregion-filling. To resolve this obstacle, we propose a screentone variationalautoencoder, ScreenVAE, to map the screened manga to an intermediate domain.This intermediate domain can summarize local texture characteristicsand is interpolative. With this domain, we effectively unify the properties ofscreening and color-filling, and ease the learning for bidirectional translationbetween screened manga and color comics. To carry out the bidirectionaltranslation, we further propose a network to learn the translation between the intermediate domain and color comics. Our model can generate qualityscreened manga given a color comic, and generate color comic that retainsthe original screening intention by the bitonal manga artist. Several resultsare shown to demonstrate the effectiveness and convenience of the proposedmethod. We also demonstrate how the intermediate domain can assist otherapplications such as manga inpainting and photo-to-comic conversion.