Component detection in Chinese character using CNN

• Item type: SIG Technical Reports(1)
• 公開日: 2017-01-12
• タイトル: Component detection in Chinese character using CNN
• タイトル / 言語: en
• タイトル: Component detection in Chinese character using CNN
• 言語: eng
• キーワード / 主題Scheme: Other
• キーワード / 主題: 一般セッション2
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: Kyushu Univerisity
• 著者所属: Kyushu Univerisity
• 著者所属: Tokyo University
• 著者所属: Kyushu Univerisity
• 著者所属(英): en
• 著者所属(英): Kyushu Univerisity
• 著者所属(英): en
• 著者所属(英): Kyushu Univerisity
• 著者所属(英): en
• 著者所属(英): Tokyo University
• 著者所属(英): en
• 著者所属(英): Kyushu Univerisity
• 著者名: Letao, Zhou; Brian, Kenji Iwana; Kumiko, Tanaka-Ishii; Seiichi, Uchida
• 著者名(英): Letao, Zhou; Brian, Kenji Iwana; Kumiko, Tanaka-Ishii; Seiichi, Uchida
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: In recent years, CNN (Convolutional Neural Network)-based models have achieved powerful results in the multi-object detection task. However, most models use region-based frameworks, which do region extraction before the CNN's training. Thus, it is still unknown if the CNN can directly understand the image's structure. This study aims to evaluate the CNN's ability of learn the structure of an image by detecting the components (sub-structure) of a character from the line-based Chinese character images.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: In recent years, CNN (Convolutional Neural Network)-based models have achieved powerful results in the multi-object detection task. However, most models use region-based frameworks, which do region extraction before the CNN's training. Thus, it is still unknown if the CNN can directly understand the image's structure. This study aims to evaluate the CNN's ability of learn the structure of an image by detecting the components (sub-structure) of a character from the line-based Chinese character images.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11131797
• 書誌情報: 研究報告コンピュータビジョンとイメージメディア（CVIM）; 巻 2017-CVIM-205, 号 8, p. 1-5, 発行日 2017-01-12
• ISSN / 収録物識別子タイプ: ISSN
• ISSN / 収録物識別子: 2188-8701
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会