@techreport{oai:ipsj.ixsq.nii.ac.jp:00242242, author = {千葉, 竜太 and 金子, 直史 and Ryuta, Chiba and Naoshi, Kaneko}, issue = {16}, month = {Jan}, note = {近年,手書き文字の生成に関する研究は盛んに行われているものの,文字の美化に関する研究はまだ少ない.本稿では,近年の文字生成で提案されている文字のスタイル特徴とコンテンツ特徴のもつれ解きほぐしに着目し,スタイル変換を用いて手書き文字の美化を行う.オリジナルの文字画像とその骨格画像をペアとして学習させることでコンテンツのみの文字画像からスタイル化された文字画像の生成を可能にする.骨格が異なる文字間でのスタイル転送と,文字の位置とサイズに依存しないスタイル転送を実験により検証し,文字の美化に応用できる可能性を示した., In recent years, research on handwritten character generation has been actively conducted, but studies specifically focusing on character beautification remain limited. This paper addresses the beautification of handwritten characters by leveraging style-content disentanglement, a concept recently proposed in character generation research. By learning from pairs of original character images and their skeletonized images, the proposed method enables the generation of stylized character images from content-only character images. We validate the method through experiments involving style transfer between characters with differing skeletons, as well as style transfer that is invariant to character position and size. The results demonstrate the potential applicability of style transfer techniques to handwritten character beautification.}, title = {スタイルとコンテンツのもつれ解きによる手書き文字の美化可能性の検討}, year = {2025} }