| Co-Investigator(Kenkyū-buntansha) |
OKABE Naoki Department of Information and Computer Engineering Professor, 情報工学科, 教授 (90109273)
YOSHIMURA Mitsu College of Business Administration and Information Science, Chubu University Pro, 経営情報学部, 教授 (60109311)
KIMURA Fumitaka Faculty of Engineering, Mie University Associate Professor, 情報工学科, 助教授 (00115560)
TSURUOKA Shinji Toyata National College of Technology Faculty of Engineering, Mie University Ass, 電子工学科, 助教授 (30126982)
TAKESHITA Tetsuo Toyata National College of Technology Department of Information and Computer Eng, 情報工学科, 教授 (20149933)
|
|---|
| Research Abstract |
(1) Authors developed a program library for document understanding and its applications. They named this library SMART, which means Subroutine package for MAchine Recognition of characTers, and registered it with the Computation Center of Nagoya University. For researchers, especially for beginners, who are interested in document understanding or character recognition, they made two manuals on how to use this library programs as follows.
Manual on Program Package SMART for Character Recognition I Subroutine & Function (160 pages)
Manual on Program Package SMART for Character Recognition II Test Program (80 pages)
(2) They studied pattern recognition theory which is the basis of the programs they developed. They studied the performance of pattern classifiers and the size of design samples and confirmed that modified classifier designed by using Stein's estimator achives better performance than the conventional ones.
(3) They also made applied researches on character recognition, zip code recognition, writer identification and signature verification to confirm the effectiveness of developed program library.
As for the character recognition, they got recognition rate by 97.74 % using handwritten Chinese characters contained in the JIS 1st level database, and 91-98 % rate in the case of the recognition experiments of unconstrained handwritten word.
As for the zip code recognition experiments, making use of localized direction histgram, they got average 99.18 % recognition rate, using as data zip codes offered by the Ministry of Posts and Telecommunications.
As for the writer recognition, using as experiments data general Japanese and Korean sentence, they got good results.
As for the signature verification study, they made experiments on online signature verification, offline signature verification and got many satisfactory results. They also investigated an automatic verification system for Japanese coutersignatures on travelers checks.
|
