Abstract
We developed "Comicolorization", a semi-automatic colorization system formanga images. Given a monochrome manga and reference images as inputs, oursystem generates a plausible color version of the manga. This is the first workto address the colorization of an entire manga title (a set of manga pages).Our method colorizes a whole page (not a single panel) semi-automatically, withthe same color for the same character across multiple panels. To colorize thetarget character by the color from the reference image, we extract a colorfeature from the reference and feed it to the colorization network to help thecolorization. Our approach employs adversarial loss to encourage the effect ofthe color features. Optionally, our tool allows users to revise thecolorization result interactively. By feeding the color features to our deepcolorization network, we accomplish colorization of the entire manga using thedesired colors for each panel.