返回
FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK

FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK

ID：38275219

大小：36.89 KB

页数：6页

时间：2019-05-27

FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK_第1页
FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK_第2页
FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK_第3页
FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK_第4页
FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK_第5页
资源描述：

《FEATURE SELECTION AND CHARACTER CLASSIFICATION USING A WEIGHTLESS ARTIFICIAL NEURAL NETWORK》由会员上传分享，免费在线阅读，更多相关内容在行业资料-天天文库。

1、FEATURESELECTIONANDCHARACTERCLASSIFICATIONUSINGAWEIGHTLESSARTIFICIALNEURALNETWORKAndreasGarzottoSwissLife,InformationSystemsResearch,8022Zurich,SwitzerlandE-mail:garzotto@swssai.uu.ch&DepartmentofComputerScience,UniversityofZurich,SwitzerlandAbstractAweightle

2、ssANN,theNAND-net,isusedforfeatureselectionandclassificationinacharacterrecognitionsystem.Whileinmanysystemsahumanselectswhichfeaturesaretobeused,ourapproachemploysanANNtochoosecandidatesfromalargesetofpotentialfeatures,whichcanbeusedfordiscriminationbetweench

3、aracterclasses.Thisisaccomplishedbyteachingthenetworkusingatrainingsetthatincludesallpotentialfeaturesasinputs.ANAND-netcaneasilybeanalysed,sinceallconnectionsarebinaryandeverynetworkcorrespondstoalogicalexpressionindisjunctivenormalform.Therefore,theinputs(f

4、eatures)onwhichthenetworkdoesnotdependcanbediscardedbecausetheyarerecognisedwithoutdifficulty.TheremainingfeaturesetisusedforretrainingofaNAND-netthatcan,subsequently,beusedforfastcharacterclassification.1IntroductionAutomaticanalysisandrecognitionofimages,obta

5、inedbydigitisingprintedtextdocuments,isaproblemthathasnotyetbeensolvedsatisfactorily.Therecognitionofisolatedcharacterbitmapsisoneofthemajorissues.Mostknownalgorithmsforthispurposeeitherusetemplatematchingorfeaturedetection.Templatematchingisappropriateifthes

6、etofcharacterstoberecognisediswellknownandtheimagesareofgoodquality,butitfailsifarbitraryfontsorstronglydistortedcharactersshouldbeclassified.Featuredetectionismuchmoreflexiblebecauseitdoesnotrelyonthedigitisedbitmapitself,butonfeaturesfoundwithinthebitmap.Thec

7、entralquestionwiththelatterapproachis:Whichfeaturesarerequiredfortheclassificationofcharacters?Inmanycases,thefeaturesareselectedbyahuman.Ahuman,however,isbiasedtowardscertainchoicesandmayoverlookotherusefulfeatures.Thispaperpresentsanapproachtoselectingfeatur

8、esfromalargesetofpotentialfeaturesautomaticallybyusingaweightlessartificialneuralnetwork(ANN).2ANNsandCharacterRecognitionThemoststraightforwardwaytoapplyanANNtocharacterrecognitionistofee

当前文档最多预览五页，下载文档查看全文

此文档下载收益归作者所有

当前文档最多预览五页，下载文档查看全文
温馨提示：
1. 部分包含数学公式或PPT动画的文件，查看预览时可能会显示错乱或异常，文件下载后无此问题，请放心下载。
2. 本文档由用户上传，版权归属用户，天天文库负责整理代发布。如果您对本文档版权有争议请及时联系客服。
3. 下载前请仔细阅读文档内容，确认文档内容符合您的需求后进行下载，若出现内容与标题不符可向本站投诉处理。
4. 下载文档时可能由于网络波动等原因无法下载或下载错误，付费完成后未能成功下载的用户请联系客服处理。