Presentation Information
[2E5-GS-10o-06]A Study on Automated Gap Filling Method for Animation Line Drawings using Deep Learning
Tomoya Murata1, 〇Naoki Mori1 (1. Osaka Metropolitan University)
Keywords:
Animation Production,Deep Learning,Image Inpainting,Line Art,Colorization Support
In recent years, animation production has become increasingly digital, but automation of the coloring process, known as the “finishing” stage, is still an important challenge. In this process, regions enclosed by line drawings must be detected, and flat coloring is applied by filling each region with a uniform color. However, when small gaps exist in the line drawings, color leakage occurs during region segmentation, and fill tools do not work correctly. As a result, artists currently spend a large amount of time fixing these gaps by hand.
Recent image generation AI models have shown high drawing ability, but they tend to reconstruct the entire image. Because of this, it is difficult for them to keep the original lines unchanged while performing strict control at the pixel level. In this study, we examine a method for filling only the gaps in line drawings using deep learning, without damaging the original lines.
The proposed method uses feature maps that represent the spatial density of lines. Based on these maps, a neural network estimates gap locations by considering both local shapes and global context. At present, we have confirmed that the method can suggest suitable candidates for filling gaps in line drawings, and we are evaluating its effectiveness as a tool for supporting animation production.
Recent image generation AI models have shown high drawing ability, but they tend to reconstruct the entire image. Because of this, it is difficult for them to keep the original lines unchanged while performing strict control at the pixel level. In this study, we examine a method for filling only the gaps in line drawings using deep learning, without damaging the original lines.
The proposed method uses feature maps that represent the spatial density of lines. Based on these maps, a neural network estimates gap locations by considering both local shapes and global context. At present, we have confirmed that the method can suggest suitable candidates for filling gaps in line drawings, and we are evaluating its effectiveness as a tool for supporting animation production.
Comment
To browse or post comments, you must log in.Log in