Colorization method of high resolution anime sketch with Pix2PixHD

Abstract

Image colorization is an important field of computer vision. With the increasing resolution of colorized images, people’s requirements for the quality of the coloring effect of pictures have been increasingly improved, and the effects of traditional image colorization methods can not longer resolve with high-resolution colorization problem. This paper proposed a colorization method for high-resolution anime sketch based on conditional generation confrontation network. By using a network model with multi-scale generator and multi-scale discriminator, the mapping relationship between the anime sketch and the corresponding image was learned and optimized in the process of generator and discriminator training. Finally, the trained network model was used to color the anime sketch. Experiment results show that compared with other anime sketch colorization methods, the proposed in this paper can color high resolution anime sketch while maintaining considerable visual effects.