cGAN-based Manga Colorization Using a Single Training Image

Paulina Hensman,Kiyoharu Aizawa

DOI: https://doi.org/10.48550/arXiv.1706.06918

2017-06-21

Abstract:The Japanese comic format known as Manga is popular all over the world. It is traditionally produced in black and white, and colorization is time consuming and costly. Automatic colorization methods generally rely on greyscale values, which are not present in manga. Furthermore, due to copyright protection, colorized manga available for training is scarce. We propose a manga colorization method based on conditional Generative Adversarial Networks (cGAN). Unlike previous cGAN approaches that use many hundreds or thousands of training images, our method requires only a single colorized reference image for training, avoiding the need of a large dataset. Colorizing manga using cGANs can produce blurry results with artifacts, and the resolution is limited. We therefore also propose a method of segmentation and color-correction to mitigate these issues. The final results are sharp, clear, and in high resolution, and stay true to the character's original color scheme.

Graphics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to color black - and - white Manga automatically. Specifically, traditionally, Japanese Manga are usually black and white, and manual coloring is both time - consuming and expensive. Existing automatic coloring methods usually rely on gray - scale values, but this information does not exist in Manga. In addition, due to copyright protection, the available color Manga data for training is very limited.

To overcome these problems, the author proposes a Manga coloring method based on conditional generative adversarial network (cGAN), which only requires one color reference image for training, thus avoiding the need for large - scale data sets. Through this method, the author hopes to generate clear, high - quality color Manga images while maintaining the original color scheme of the characters.

### Main problem summary:
1. **Lack of gray - scale information**: Traditional automatic coloring methods rely on gray - scale values, but Manga is black and white and has no gray - scale information.
2. **Copyright issues**: Due to copyright protection, the available color Manga training data is very scarce.
3. **Time and cost**: Manual coloring of Manga requires a great deal of time and professional skills and is costly.

### Solutions:
- **Single training image**: Use only one color reference image to train cGAN, reducing the dependence on large - scale data sets.
- **Segmentation and color correction**: To solve the problems of blurring and artifacts that may occur during the cGAN coloring process, the methods of segmentation and color correction are proposed.
- **High - resolution output**: Ensure that the final result is clear, high - quality, and faithful to the original color scheme of the characters.

Through these methods, the author hopes to simplify the Manga coloring process, reduce human intervention, and improve the quality and efficiency of coloring.