{"created":"2023-05-15T14:47:24.131302+00:00","id":2559,"links":{},"metadata":{"_buckets":{"deposit":"95136793-c186-4815-a14b-b6df2e79a263"},"_deposit":{"created_by":3,"id":"2559","owners":[3],"pid":{"revision_id":0,"type":"depid","value":"2559"},"status":"published"},"_oai":{"id":"oai:rekihaku.repo.nii.ac.jp:00002559","sets":["21:246"]},"author_link":["6254","6255"],"control_number":"2559","item_10002_biblio_info_7":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicIssueDates":{"bibliographicIssueDate":"2020-03-31","bibliographicIssueDateType":"Issued"},"bibliographicPageEnd":"75","bibliographicPageStart":"47","bibliographicVolumeNumber":"220","bibliographic_titles":[{"bibliographic_title":"国立歴史民俗博物館研究報告","bibliographic_titleLang":"ja"},{"bibliographic_title":"Bulletin of the National Museum of Japanese History","bibliographic_titleLang":"en"}]}]},"item_10002_description_19":{"attribute_name":"フォーマット","attribute_value_mlt":[{"subitem_description":"application/pdf","subitem_description_type":"Other"}]},"item_10002_description_5":{"attribute_name":"抄録","attribute_value_mlt":[{"subitem_description":"本研究では，デジタルアーカイブ画像のメタデータを生成し類似画像検索などに役立てることを目的にしている。\n一般物体認識でよく用いられている画像のヒストグラム表現手法，Bag-of-Features［4］ではSIFT [2] [3]に代表される画素の濃淡分布をもとに算出された特徴点および局所特徴量が用いられるが，その一方で一般物体認識の分野でDeep Learningを用いた技術［6］が注目を集めている。Deep Learning手法では，画像全体を入力し，画像中に存在する主となる物体を認識させることが一般的となっており，画像中の様々な局所的な情報が欠落してしまっていた。\nそこで本研究では画像をセグメントに分割し，各セグメントからDeep Learningを用いた特徴抽出を行い，クラスタリングによって分類された各セグメントのクラスタ情報を局所特徴としたBag-of-Featuresを行い，ヒストグラム表現とすることで画像に存在する意味情報を反映したメタデータ生成を提案する。また，ヒストグラム間の比較にはクラスタ間の類似関係を反映した距離計算を行うことでクラスタ数が細かすぎる際に，似ている画像が類似画像として判定できない問題を解決した。\n実験では，デジタルアーカイブとして小袖屛風画像［9］を用いてヒストグラム間の比較を行うことでDeep Learning［7］を用いてBag-of-Featuresの応用を行うことの有効性，さらにクラスタ間の距離関係を反映した距離計算を行うことの有効性を示した。","subitem_description_type":"Abstract"},{"subitem_description":"The goal of this study is to generate \"metadata\" of the images in digital archives and to use them for similar image retrieval system. Generally, bag-of-features [4], which is a histogram representation method for object recognition in images, uses feature points and local features calculated based on the grayscale distribution of pixels by using SIFT [3]. On the other hand, the deep learning approach [6] has attracted attention in the field of general object recognition as end-to-end learning, not local features oriented learning. The typical deep learning recognizes the whole structured object in the image; however, it misses significant sub-parts in the image. To overcome the issue, we divide an image into segments, extract features from each segment using deep learning, then apply bag-of-features using the clustering for the local features, and finally represents it as histogram expression reflects the metadata in the image. Furthermore, by using distance calculation representing the similarity of the cluster as comparing the histograms, the discriminant precision of a similar image could be improved, when the number of clusters is too small. The experiment result shows the effectiveness of generating metadata using BoF with deep learning, the distance evaluation method reflecting the relationship between clusters.","subitem_description_type":"Abstract"}]},"item_10002_heading_23":{"attribute_name":"見出し","attribute_value_mlt":[{"subitem_heading_banner_headline":"[共同研究] 歴史資料デジタルアーカイブデータを用いた知的構造の創生に関する研究 : 小袖屛風を対象として","subitem_heading_language":"ja"},{"subitem_heading_banner_headline":"[Collaborative Research] Research on the Creation of Intellectual Structure Using Digital Archive Data of Historical Materials : Focusing on Kosode Folding Screen","subitem_heading_language":"en"}]},"item_10002_publisher_8":{"attribute_name":"出版者","attribute_value_mlt":[{"subitem_publisher":"国立歴史民俗博物館","subitem_publisher_language":"ja"}]},"item_10002_relation_17":{"attribute_name":"関連サイト","attribute_value_mlt":[{"subitem_relation_name":[{"subitem_relation_name_text":"第220集 収録論文 タイトルリスト"}],"subitem_relation_type_id":{"subitem_relation_type_id_text":"https://www.rekihaku.ac.jp/outline/publication/ronbun/ronbun9/index.html#no220","subitem_relation_type_select":"URI"}}]},"item_10002_source_id_11":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN00377607","subitem_source_identifier_type":"NCID"}]},"item_10002_source_id_9":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"0286-7400","subitem_source_identifier_type":"PISSN"}]},"item_10002_version_type_20":{"attribute_name":"著者版フラグ","attribute_value_mlt":[{"subitem_version_resource":"http://purl.org/coar/version/c_970fb48d4fbd8a85","subitem_version_type":"VoR"}]},"item_creator":{"attribute_name":"著者","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"濱上, 知樹","creatorNameLang":"ja"},{"creatorName":"ハマガミ, トモキ","creatorNameLang":"ja-Kana"}],"nameIdentifiers":[{}]},{"creatorNames":[{"creatorName":"Hamagami, Tomoki","creatorNameLang":"en"}],"nameIdentifiers":[{}]}]},"item_files":{"attribute_name":"ファイル情報","attribute_type":"file","attribute_value_mlt":[{"accessrole":"open_date","date":[{"dateType":"Available","dateValue":"2021-03-31"}],"displaytype":"detail","filename":"kenkyuhokoku_220_04.pdf","filesize":[{"value":"4.8 MB"}],"format":"application/pdf","licensetype":"license_note","mimetype":"application/pdf","url":{"label":"kenkyuhokoku_220_04.pdf","url":"https://rekihaku.repo.nii.ac.jp/record/2559/files/kenkyuhokoku_220_04.pdf"},"version_id":"e0c3570e-7f20-4690-b8d4-84aed66416ca"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"デジタルアーカイブ","subitem_subject_scheme":"Other"},{"subitem_subject":"Deep Learning","subitem_subject_scheme":"Other"},{"subitem_subject":"一般物体認識","subitem_subject_scheme":"Other"},{"subitem_subject":"BoF","subitem_subject_scheme":"Other"},{"subitem_subject":"類似画像検索","subitem_subject_scheme":"Other"},{"subitem_subject":"deep learning","subitem_subject_language":"en","subitem_subject_scheme":"Other"},{"subitem_subject":"general object recognition","subitem_subject_language":"en","subitem_subject_scheme":"Other"},{"subitem_subject":"bug of feature","subitem_subject_language":"en","subitem_subject_scheme":"Other"},{"subitem_subject":"similar image retrieval","subitem_subject_language":"en","subitem_subject_scheme":"Other"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourcetype":"departmental bulletin paper","resourceuri":"http://purl.org/coar/resource_type/c_6501"}]},"item_title":"デジタルアーカイブにおける Deep Learning を用いたメタデータ生成","item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"デジタルアーカイブにおける Deep Learning を用いたメタデータ生成","subitem_title_language":"ja"},{"subitem_title":"Metadata Generation Using Deep-Learning in Digital Archives","subitem_title_language":"en"}]},"item_type_id":"10002","owner":"3","path":["246"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2021-03-31"},"publish_date":"2021-03-31","publish_status":"0","recid":"2559","relation_version_is_last":true,"title":["デジタルアーカイブにおける Deep Learning を用いたメタデータ生成"],"weko_creator_id":"3","weko_shared_id":-1},"updated":"2023-08-16T04:48:20.403867+00:00"}