Quad-Tree based Image Encoding Methods for Data-Adaptive Visual Feature Learning

• Item type: SIG Technical Reports(1)
• 公開日: 2015-05-11
• タイトル: Quad-Tree based Image Encoding Methods for Data-Adaptive Visual Feature Learning
• タイトル / 言語: en
• タイトル: Quad-Tree based Image Encoding Methods for Data-Adaptive Visual Feature Learning
• 言語: eng
• 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: IST, Graduate School of Informatics, Kyoto University
• 著者所属: IST, Graduate School of Informatics, Kyoto University
• 著者所属: IST, Graduate School of Informatics, Kyoto University
• 著者所属(英): en
• 著者所属(英): IST, Graduate School of Informatics, Kyoto University
• 著者所属(英): en
• 著者所属(英): IST, Graduate School of Informatics, Kyoto University
• 著者所属(英): en
• 著者所属(英): IST, Graduate School of Informatics, Kyoto University
• 著者名: Cuicui, Zhang; Xuefeng, Liang; Takashi, Matsuyama
• 著者名(英): Cuicui, Zhang; Xuefeng, Liang; Takashi, Matsuyama
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Visual feature learning is fundamental to many computer vision tasks. State-of-art methods adopt an image block based multilayer framework to learn hierarchical feature representations. However, the image block is not adaptive for low-level feature extraction and the image pyramid based hierarchical models are neither adaptive nor flexible enough to learn high-level features. To solve these problems, this thesis exploits the image spatial and hierarchical structure using Quad-Trees and employs them for local feature analysis and for hierarchical feature learning. To evaluate the reliability of our methods, we also conduct feature learning in other challenging situations: feature learning with small training data and feature learning in dynamic environments (moving camera videos). Face recognition and motion segmentation are utilized as research backgrounds for algorithm evaluation. Experimental results demonstrate the effectiveness of our methods.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Visual feature learning is fundamental to many computer vision tasks. State-of-art methods adopt an image block based multilayer framework to learn hierarchical feature representations. However, the image block is not adaptive for low-level feature extraction and the image pyramid based hierarchical models are neither adaptive nor flexible enough to learn high-level features. To solve these problems, this thesis exploits the image spatial and hierarchical structure using Quad-Trees and employs them for local feature analysis and for hierarchical feature learning. To evaluate the reliability of our methods, we also conduct feature learning in other challenging situations: feature learning with small training data and feature learning in dynamic environments (moving camera videos). Face recognition and motion segmentation are utilized as research backgrounds for algorithm evaluation. Experimental results demonstrate the effectiveness of our methods.
• 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 2015-CVIM-197, 号 33, p. 1-16, 発行日 2015-05-11
• 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