标题: 具有个人调适功能之手写中文辨识系统
A Personal Adaptive Module for Handwritten Chinese Charactes Recognition Syatem
作者: 庄舜清
Chuang, Soon-Ching
傅心家
Fu Hsin-Chia
资讯科学与工程研究所
关键字: 调适;离线;手写;事前资讯;辨识;adaptive;off-line;hardwritten;prior information;recognition
公开日期: 1995
摘要: This thesis presents an application of statistics theory on
off-line adaptivemodule for handwritten Chinese characters
recognition. The proposed methodconsists of three
components:(1).prior information,(2).feature selection,(3).
adaptive module recognition. In order to evaluate the proposed
recognition system,we choose 5401 fre-quently used Chinese
characters as our domain. The database of each testingand
training sample character for the original 5401 classifier was
created bythe Computer and Communication Laboratory of
Industrial Technology ResearchInstitute. And we select the most
300 frequently used Chinese characterswhich were written by five
members of our lab for ten times as the testingand training for
the adaptive module. Because the samples for the adaptivemodule
were not sepcified,our recognition system could reach a high
generalityand user-independence. Experimental results show that,
the method improvesrecognition rate from 44.09% to 90.03%.
本论文的目的是应用统计理论于个人手写中文字之调适。离线手写
中 文字辨识多年来一直是文字辨识的重要问题,其困难的原因是由于不
同的 使用者之间字迹变异度很大,以致难以只靠资料库的统计特征而达
到满意 的正确分类。本研究事先利用资料库所提供之〝事前资讯〞,再
配合每个 使用者个别的手写字迹特性,来建构其个人专属之手写中文字
调适模组, 以达成提升辨识率的目的。本个人手写调适模组包括三个主
要的部份,分 别是(1)个人手写字集事前资讯(prior information)之建
立,(2)特征撷取 ,(3)以模组化个人调适器完成辨识。本论文所能辨识
的字为教育部所选 定的5401个常用字,并从国小课本选定之六百零五个
常用字中隔字挑出三 百个字,再分别请五位使用者分别撰写十次,作为
训练及测试之样本来进 行实验。根据所提方法实验之结果,藉由此个人
手写调适模组的加入,系 统的整体正确率平均由 44.09% 逐渐提升至
90.03% 。因测试者不限定某 特定个别使用者,因此所建立之系统具有
广泛性及一般性。
URI: http://140.113.39.130/cdrfb3/record/nctu/#NT840392071
http://hdl.handle.net/11536/60418
显示于类别:Thesis