Project/Area Number |
61460231
|
Research Category |
Grant-in-Aid for General Scientific Research (B)
|
Allocation Type | Single-year Grants |
Research Field |
Informatics
|
Research Institution | TOHOKU UNIVERSITY |
Principal Investigator |
KIMURA masayuki Professor Faculty of Engineering Tohoku University, 工学部, 教授 (50006219)
|
Co-Investigator(Kenkyū-buntansha) |
ABE Masato Assistant Reserch Center for Applied Information Science Tohoku University, 応用情報学研究センター, 助手 (00159443)
菰田 保男 (菰田 保夫) 東北大学, 工学部, 助手 (30108469)
EJIMA Toshiaki Associate Professor The Technological University of Nagaoka, 助教授 (00124553)
KOMOTA Yasuo Assistant Faculty of Engineering Tohoku University (KAWAZOE,Yosh)
|
Project Period (FY) |
1986 – 1987
|
Project Status |
Completed (Fiscal Year 1987)
|
Budget Amount *help |
¥5,700,000 (Direct Cost: ¥5,700,000)
Fiscal Year 1987: ¥1,100,000 (Direct Cost: ¥1,100,000)
Fiscal Year 1986: ¥4,600,000 (Direct Cost: ¥4,600,000)
|
Keywords | Character extraction / Character recognition / Hand wretten character / マルチテンプレート法 / 構造解析 / 活字認識 / 手書き文字認識 / 漢字認識 / 文字連接情報 / 文字自動切出 / 単語辞書 |
Research Abstract |
The results obtained in the research project are summerized as follows. 1.Automatic extraction of characters we have developed an useful technique which makes it possible to extract characters one by one automatically from a text without restrictions on its format. The technique consists of two parts. The first part is concerned with grasping the global structure of the text from the histograms of pixels in character-images that are obtained by scanning in both the horizontal and vertical ways through out the text.The second part is concerned with finding the local structure of the text based on the histograms and finally extracting characters one by one. 2. Printed character recognition An useful techiqus of printed character recognition withhigh speed and high accuracy has been developed. The technique consists of a pattern matching method,which we shall call as an associative pattern matching, for rough classification and method of structural analysis. The accumulated percentage accuracy up to the 3rd candidates was 100% for rough classification of 17790 samples where 17790 samples=2965(kinds of chainese characters)x6(samples/kind)and that up the lst candidata was 99.99% as a result of the structural analysis based on the rough classification. 3.Hand written character recognition We have derived an technique for rough classification of hand witten characters which consists of two stages of rough classifications.The lst stage is the associative pattern matching method and the 2nd stage uses an multi-templates method.The accumulated percentage accuracy up to 4th candidate was over 99% for 60720 samples where 60720= 759(kinds of characters)x80(samples/kind).
|