Propose Character Generation in Handwriting Feature Extraction using Variational AutoEncoder

カテゴリ: 部門大会

論文No: SS3-3

グループ名: 【C】平成29年電気学会電子・情報・システム部門大会講演論文集

発行日: 2017/09/06

タイトル(英語): Propose Character Generation in Handwriting Feature Extraction using Variational AutoEncoder

著者名: 山田 智輝(岐阜大学),細江 麻梨子(岐阜県警察),加藤 邦人(岐阜大学),山本 和彦(岐阜大学)

著者名(英語): Tomoki Yamada|Mariko Hosoe|Kunihito Kato|Kazuhiko Yamamoto

キーワード: 文字生成|生成モデル|ディープラーニングディープラーニング|character generation|generative model|deep learning

要約(日本語): Handwriting analysis is a method to determine an unknown writer by comparing known writer’s characters with unknown writer’s characters. The lack of characters of the same class is one of problems with the accuracy of handwriting analysis. We tried to generate different class of characters possessing handwriting features to solve the lack of characters. In this paper, we tried the learning method using deep learning to model the handwriting feature extraction and the handwriting character data generative model. We tried to learn the generative model of the data of Japanese character (Hiragana). In this paper, we propose a modified method to improve character generation accuracy by adding a convolution model to existing neural network model.

PDFファイルサイズ: 562 Kバイト