著者
青池 亨 里見 航 川島 隆徳
雑誌
じんもんこん2018論文集
巻号頁・発行日
vol.2018, pp.97-102, 2018-11-24

The National Diet Library is now developing techniques for automatically recognizing which areas of a printed page are illustrations and which are graphemes, as a means of improving the searchability of digitized material. The ability to distinguish between illustrations and graphemes is expected to im- prove the accuracy of OCR processing by allowing areas without graphemes to be ignored while ena- bling the application of contrast correction to areas with graphemes, thereby improving readability of the digital images. Moreover, the ability to extract areas with illustrations is expected to have practical applications for content-based retrieval of similar images. This paper focuses on the extraction of areas with illustrations and reports on the creation of a system that is consistently able to extract illustra- tions from digital images of documents as well as perform content-based retrieval of images.Services incorporating these proposed techniques will be released on a trial basis on the NDL Lab web- site. (https://lab.ndl.go.jp/).

言及状況

Twitter (53 users, 53 posts, 47 favorites)

NDLラボ「次世代デジタルライブラリー」https://t.co/ln38PMe5SV じんもんこん2018「資料画像中の挿絵領域の自動抽出及び画像検索システムの実装」https://t.co/iqgh5sHFGA で「平成30年度内を目途に公開予定」とされていた検索システムがついに公開! https://t.co/SqWhpQNdP5

収集済み URL リスト