Interactive segmentation for manga using lossless thinning and coarse labeling

Yuji Aramaki, Yusuke Matsui, Toshihiko Yamasaki, Kiyoharu Aizawa

Abstract: Manga (Japanese comics) are popular all over the world. Applications using manga, such as automatic layout suggestion and character composition, have been proposed recently. To implement these applications, important but time-consuming manga segmentation is required. We propose a scribble-based interactive segmentation method for manga images. Based on Smart Scribbles, we propose (1) lossless thinning, which enables the handling of complex raster images such as manga, and (2) coarse labeling, which drastically reduces the computational cost of labeling and enables accurate segmentation in real time. Our experimental results show that the proposed method efficiently produces much more accurate results than does the Quick Selection Tool in Adobe Photoshop1 <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">®</sup> or the connected-component labeling method.