{"links":{},"id":2007003,"metadata":{"_oai":{"id":"oai:ipsj.ixsq.nii.ac.jp:02007003","sets":["581:1765244505933:1765246169608"]},"path":["1765246169608"],"owner":"80578","recid":"2007003","title":["深層学習による削屑木簡の画像復元"],"pubdate":{"attribute_name":"PubDate","attribute_value":"2026-02-15"},"_buckets":{"deposit":"2c1e0dc2-4f4e-4b30-bf7e-c4f2daf1feaf"},"_deposit":{"id":"2007003","pid":{"type":"depid","value":"2007003","revision_id":0},"owner":"80578","owners":[80578],"status":"published","created_by":80578},"item_title":"深層学習による削屑木簡の画像復元","author_link":[],"item_titles":{"attribute_name":"タイトル","attribute_value_mlt":[{"subitem_title":"深層学習による削屑木簡の画像復元","subitem_title_language":"ja"},{"subitem_title":"Image-based Restoration of Mokkan Fragments Using Deep Learning","subitem_title_language":"en"}]},"item_keyword":{"attribute_name":"キーワード","attribute_value_mlt":[{"subitem_subject":"[特集：人文科学とコンピュータ] 木簡，画像復元，敵対的生成ネットワーク，拡散モデル","subitem_subject_scheme":"Other"}]},"item_type_id":"2","publish_date":"2026-02-15","item_2_text_3":{"attribute_name":"著者所属","attribute_value_mlt":[{"subitem_text_value":"東京電機大学"},{"subitem_text_value":"東京電機大学"}]},"item_2_text_4":{"attribute_name":"著者所属(英)","attribute_value_mlt":[{"subitem_text_value":"Tokyo Denki University","subitem_text_language":"en"},{"subitem_text_value":"Tokyo Denki University","subitem_text_language":"en"}]},"item_language":{"attribute_name":"言語","attribute_value_mlt":[{"subitem_language":"jpn"}]},"publish_status":"0","weko_shared_id":-1,"item_file_price":{"attribute_name":"Billing file","attribute_type":"file","attribute_value_mlt":[{"url":{"url":"https://ipsj.ixsq.nii.ac.jp/record/2007003/files/IPSJ-JNL6702012.pdf","label":"IPSJ-JNL6702012.pdf"},"date":[{"dateType":"Available","dateValue":"2028-02-15"}],"format":"application/pdf","billing":["billing_file"],"filename":"IPSJ-JNL6702012.pdf","filesize":[{"value":"12.7 MB"}],"mimetype":"application/pdf","priceinfo":[{"tax":["include_tax"],"price":"660","billingrole":"5"},{"tax":["include_tax"],"price":"330","billingrole":"6"},{"tax":["include_tax"],"price":"0","billingrole":"8"},{"tax":["include_tax"],"price":"0","billingrole":"44"}],"accessrole":"open_date","version_id":"69e07abb-a281-456a-b824-9d9edf19640b","displaytype":"detail","licensetype":"license_note","license_note":"Copyright (c) 2026 by the Information Processing Society of Japan"}]},"item_2_creator_5":{"attribute_name":"著者名","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"塩野,健治"}]},{"creatorNames":[{"creatorName":"大山,航"}]}]},"item_2_creator_6":{"attribute_name":"著者名(英)","attribute_type":"creator","attribute_value_mlt":[{"creatorNames":[{"creatorName":"Kenji Shiono","creatorNameLang":"en"}]},{"creatorNames":[{"creatorName":"Wataru Ohyama","creatorNameLang":"en"}]}]},"item_2_source_id_9":{"attribute_name":"書誌レコードID","attribute_value_mlt":[{"subitem_source_identifier":"AN00116647","subitem_source_identifier_type":"NCID"}]},"item_resource_type":{"attribute_name":"資源タイプ","attribute_value_mlt":[{"resourceuri":"http://purl.org/coar/resource_type/c_6501","resourcetype":"journal article"}]},"item_2_publisher_15":{"attribute_name":"公開者","attribute_value_mlt":[{"subitem_publisher":"情報処理学会","subitem_publisher_language":"ja"}]},"item_2_source_id_11":{"attribute_name":"ISSN","attribute_value_mlt":[{"subitem_source_identifier":"1882-7764","subitem_source_identifier_type":"ISSN"}]},"item_2_description_7":{"attribute_name":"論文抄録","attribute_value_mlt":[{"subitem_description":"重要な記述史料の一種である木簡は，表面を刀子で削り取り再利用されていたため，削り出された削屑木簡が大量に出土している．これらの削り屑木簡には文字や墨痕が残されているが，その多くは欠損しており，翻刻が困難である．削り屑木簡の欠損部位をデジタル画像上で復元できれば，木簡文字翻刻の支援となると著者らは考える．本研究では，代表的な画像生成深層ニューラルネットワークである敵対的生成ネットワーク（Generative Adversarial Network: GAN）と拡散モデル（Diffusion Model: DM）のそれぞれを用いた木簡欠損部位の画像復元を提案し，それらの性能を評価した．モデルの学習，評価のために，奈良文化財研究所提供の木簡画像データセットを構築し，画質評価（PSNR，SSIM，FID）および検索精度を指標にして提案手法の有効性を検証した．検証の結果，GAN，DMはそれぞれ，視覚的な自然さ，文字や墨痕の構造的再現性に優れた復元を実現できることが示された．復元画像を用いた画像検索では，非復元画像よりも高精度な画像検索が可能となり，画像復元の木簡研究への貢献が期待される．","subitem_description_type":"Other"}]},"item_2_description_8":{"attribute_name":"論文抄録(英)","attribute_value_mlt":[{"subitem_description":"Mokkan, ancient wooden tablets inscribed with characters, are vital historical artifacts. However, many mokkan were scraped and reused, resulting in fragmented pieces (mokkan fragments) that retain only partial ink traces, making transcription difficult. This study proposes image inpainting methods using deep learning to restore missing regions of mokkan images digitally. We implement two approaches: one based on a Generative Adversarial Network (GAN) with edge-guided reconstruction and another using a Diffusion Model (DM) guided by structural features. A dataset combining mokkan character images and realistic damage masks was constructed for training and evaluation. Quantitative experiments using PSNR, SSIM, and FID metrics showed that GANs produce smoother, visually consistent restorations, while DMs better preserve structural features. Additionally, image retrieval experiments using VGG16 demonstrated improved retrieval accuracy when using restored images, especially those generated by DMs. Qualitative evaluations by mokkan researchers further confirmed the utility of the restored images in historical research. Deep learning-based restoration methods can significantly aid the study and transcription of damaged mokkan. Future work includes enhancing usability through interactive tools and offering multiple restoration candidates per character to support expert analysis.","subitem_description_type":"Other"}]},"item_2_biblio_info_10":{"attribute_name":"書誌情報","attribute_value_mlt":[{"bibliographicPageEnd":"282","bibliographic_titles":[{"bibliographic_title":"情報処理学会論文誌"}],"bibliographicPageStart":"274","bibliographicIssueDates":{"bibliographicIssueDate":"2026-02-15","bibliographicIssueDateType":"Issued"},"bibliographicIssueNumber":"2","bibliographicVolumeNumber":"67"}]},"relation_version_is_last":true,"item_2_identifier_registration":{"attribute_name":"ID登録","attribute_value_mlt":[{"subitem_identifier_reg_text":"10.20729/0002007003","subitem_identifier_reg_type":"JaLC"}]},"weko_creator_id":"80578"},"created":"2026-02-09T04:16:50.394271+00:00","updated":"2026-02-16T00:34:24.024197+00:00"}