請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68309
標題: | 應用霍普菲爾德神經網路模型識別手寫字符 Handwritten Character Recognition Loading in Hopfield Model |
作者: | LING-QI ZENG 曾令祺 |
指導教授: | 劉長遠(CHANG-YUAN LIU) |
關鍵字: | 手寫識別,彎曲橢圓特徵,霍氏模型,相似性量度,模式識別,應用, Handwriting recognition,Pattern recognition,Hopfield model,Similarity measurements Bended-ellipse feature, |
出版年 : | 2017 |
學位: | 碩士 |
摘要: | 現如今,手寫字符識別系統的普及率越來越高,應用的領域也越來越豐富。橢圓彎曲特 徵模型這合表現複雜字符的幾何型態,這些特徵點在幾何上擁有許多規則,我們可以應用霍 氏模型記憶這些規則,然後達成特徵點的配對。為了 升手寫字符的識別率,我們能設計更 多不同的幾何規則,於此同時,也能解決更多不同類型的手寫字符識別問題。
在這篇論文中,我們為手寫字體的識別設計了兩項新的相似性量度方法:內特徵相似度 和內連接相似度。然後將原有的相似性量度和新的相似性量度方法結合,再應用到霍氏模型 的方法中,以求得到更好的配對結果。除了手寫字符識別系統以外,我們也 出許多符合現 今趨勢的新應用,比如音樂樂譜的識別、車輛牌照的識別和相似輪廓線間不同之處的識別等。 將本論文所 供的方法繼續發展,能夠解決實際生活中更復雜的應用。設計更多不同的幾何 規則能 升識別率,也能解抉許多不同類型的型態辨識問題。 Nowadays, handwriting recognition systems has been applied to all aspects of industries and social life. Each feature of a radical or a pattern is represented by a five dimensional vector called Bended-ellipse feature which including the coordinates the direction, the angle, and the lengths information. Once the features are generated, they will find the topological relations between those features. We simply obtain a feature-to-feature (FTF) order by define the “neighbor” of the features. After obtaining bended-ellipse features and FTF order information, we can begin the classification. It is achieved by measuring the compatibility of every radical with the handwriting pattern and standard pattern. The standard pattern which minimizes the dis-similarity is the classification result. The original feature-to-feature adhesion method uses only two similarities for classification. There are some other relations of features can be included for more accurate match. In this paper, we improve the method by making some changes to original rules and adding new similarities among features. We also give some other applications of this method by reuse the calculated similarities. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/68309 |
DOI: | 10.6342/NTU201704208 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊網路與多媒體研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-106-1.pdf 目前未授權公開取用 | 1.45 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。