miwo「みを」AIくずし字認識アプリ

Center for Open Data in the Humanities (CODH)

Mayelana nalolu hlelo lokusebenza

Ngifuna ukufunda izinto zokubhala ezihlanganisayo! Kepha angikwazi ukuyifunda! "Mio" isicelo esisiza abantu abanjalo. Uma uthatha isithombe salokho ngekhamera bese ucindezela inkinobho, i-AI izoguqula izinhlamvu ezingenamsoco zibe izinhlamvu zesimanje. Uyemukelwa emhlabeni wesikripthi esihlanganisayo.

I-ROIS-DS Humanities Open Data Shared Use Center (i-CODH) isungule ubuchwepheshe bokuqashelwa kwezinhlamvu ze-AI obunomsebenzi wokuguqula (ukuphrinta) izinhlamvu zokushefa eziqukethwe ezithombeni zibe izinhlamvu zesimanje. Uhlelo lokusebenza lwe-AI kuzuji lokuqashelwa kwe-smartphone "miwo" lwasungulwa ngenhloso yokwenza lobu buchwepheshe bube lula kunoma ngubani ukuthi abusebenzise.
"UMio" uqanjwe ngegama elithi "Mio Tsukushi", incwadi ye-14 ka "Genji Monogatari". Njengoba nje i- "Mio Tsukushi" isebenza njengesiqondisi somshayeli wabantu, sihlose ukusebenzisa uhlelo lokusebenza lwe- "Mio" njengesiqondisi sokuhamba olwandle lwezinto zokubhala ezihlanganisiwe.


[Kunconyelwe abantu abanje]
● Labo abafuna ukufunda imibhalo esetshenzisiwe esesandleni
● Labo abafuna ukusheshe bahlole okuqukethwe yizinhlamvu zeskripthi
● Labo abafuna ukufunda umbhalo ohlanganisiwe
● Labo abafuna ukubuka kuqala kokushicilelwa kabusha kombhalo ohlanganisiwe

[Ongakwenza ngohlelo lokusebenza lwe-miwo]
Kungenzeka ukubona izinhlamvu ezingenamsoco bese uziguqula (uziphrinte) zibe izinhlamvu zesimanje zezithombe ezithathwe ngekhamera noma izithombe ezilandwe kwi-Intanethi.
Izinhlamvu ezaziwayo zingaboniswa esithombeni. Ungakhombisa futhi unxande omele ukuma kwezinhlamvu.
Uma umphumela wokubonwa ungalungile, ungalungisa izinhlamvu.
Ungakhipha umphumela wokubonwa njengombhalo, uwukopishe, bese uwusebenzisa kwezinye izinhlelo zokusebenza.
Ungagcina umphumela wokuqashelwa kuhlelo lokusebenza bese uwukhumbula ngokuhamba kwesikhathi.
Ungasesha izibonelo zesikripthi esizihlanganisayo ngaphakathi kohlelo lokusebenza ngokuxhumanisa nesethi yedatha yeskripthi se-CODH.

I-Twitter: https://twitter.com/rois_codh

【Amanothi】
-Njengoba i-AI yohlelo lokusebenza lwe- "Mio" ifunda idatha ye-junk-character eqoqwe kusukela ekuphrintweni kwesikhathi se-Edo, ukunemba kwemibhalo yesikhathi se-Edo kuphakeme kakhulu, kepha izinto zezinye izikhathi, imibhalo yesandla, njll. Okwemibhalo yakudala, njll., Ukunemba kungancipha.
AccuracyUkunemba kungancipha ngokuya ngesimo sezinto ezifana namabala, okudliwe yizikelemu, amaphethini wephepha, nemvelo yokudubula efana nokukhanyisa nezithunzi.
Okwamanje, asikwazi ukubona izinhlamvu ezi-jumbled ezibhalwe kuzikhumbuzo zamatshe nakumabhodi ezimpawu.
-Lolu hlelo lokusebenza aludingi ukubhaliswa komsebenzisi futhi aluqoqi imininingwane yomuntu. Ngaphezu kwalokho, asiqoqi imininingwane yokuhlonza ngakunye yamatheminali.
--Umfanekiso olayishwe kusuka kulolu hlelo lokusebenza uye kuseva ngokuqashelwa kwezinhlamvu ze-kanji nemiphumela yokubonwa ayigcinwa kuseva.
ContentOkuqukethwe kwedokhumenti okususwe ukubethelwa kusetshenziswa lolu hlelo lokusebenza kungaqukatha imininingwane mayelana nobumfihlo bomuntu siqu, ngakho-ke sicela uhloniphe amalungelo abantu besithathu lapho babelana / beshicilela okuqukethwe.

Ngeminye imininingwane ngohlelo lokusebenza lwe- "Mio", sicela ubone http://codh.rois.ac.jp/miwo/. Uma unezicelo, imibuzo, noma izinkinga, sicela uxhumane nathi ekhelini le-imeyili elingezansi.

miwo (at) nii.ac.jp

Yini entsha

新規開発したAIくずし字認識システムRURI（瑠璃）を利用します。従来のAIくずし字認識モデルと比較して、RURIにはAI物体検出技術をくずし字認識に最適化するための工夫が加えられており、文字背景の色や模様が複雑な場合でも精度が向上しています。また国文学研究資料館が作成しCODHが公開する、100万文字以上の「日本古典籍くずし字データセット」を独自に強化することで、これまで認識できなかった文字が新たに認識できるようになりました。