[4P3-J-10] Vision, speech: image recognition learning

[4P3-J-10-03] Study on the method creating of dataset based on Grabcut and detection of white-eye region in automated colorizing grayscale line drawing of a person

〇Masashi AIZAWA¹, Yuichi SEI¹, Yasuyuki TAHARA¹, Akihiko OHSUGA¹ (1. The University of Electro-Communications)

Keywords:Automated Colorization, White-eye Region, Line Drawing

The colorizing of line drawings requires special skills, experience and knowledge. So artists need a lot of time and effort.Such a background, studies of automatic colorizing for line drawings have attracted attention in recent years.However, existing approaches have challenges. One of them is the problem that inconsistency occurs in the white-eye region of the grayscale line drawing of a person and the automatic coloring result. One of them is that inconsistency arise in white-eye region of grayscale line drawings of a person and automatic coloring results. In grayscale line drawing of a person, the region of the skin and the white-eye of a person is often expressed in white in both cases, so that existing approach often can not determine its boundary. Therefore, in this research, in addition to improving concerning reduction of burden of annotation attachment in preparation of dataset of white-eye region detection method of existing grayscale line drawing of a person,segmentation is performed by detecting white-eye region using machine learning. As a result, accuracy of white-eye region detection was reduced, but this approach succeeded in shortening the time required to create dataset.