Semi-Auto Sketch Colorization Based on Conditional Generative Adversarial Networks

The sketch plays an important role in the animation industry. Auto or semi-auto colorizing sketches will improve animators' efficiency and reduce the production costs. In this paper, we propose a semi-auto sketch colorization method based on conditional generative adversarial networks, which can support user interaction by adding scribbles to guide the colorization process. In addition, we apply a pre-model to extract high-level features of sketches in order to make good use of sketches' unique texture information. Furthermore, the loss function in our method is specially designed that can reduce blend and overflow in the result. At last, we use joint bilateral filter to smooth the output and generate a cleaner and vivid coloring sketch. Experimental results show that every module in our method can make a contribution to the final results. Moreover, the comparison with PaintsChainer demonstrates that our method can avoid large areas of leakage in the background and have cleaner skin for characters.

I. Introduction

Grayscale image colorization has been a classic problem since the invention of photography. The missing of images' color leads to less information and visual impact, so hand colorization gradually emerges. However, auto or semi-auto colorization technologies are urgently demanded due to the complexity and high requirements of hand colorization as well as the increasing number of images. Several classic methods have been proposed in the last few years.