印刷體漢字識別系統(tǒng)的特征提取和匹配識別研究.pdf

1、大連理工大學碩士學位論文印刷體漢字識別系統(tǒng)的特征提取和匹配識別研究姓名：聶玖星申請學位級別：碩士專業(yè)：信號與信息處理指導教師：李建華20081201大連理工大學碩士學位論文ResearchonFeatureExtractionandMatchingRecognitionofPrintedChineseCharacterRecognitionSystemAbstractChinesecharacterusedbymostpeopleint

2、heworldwiththousandsyearslnhistoryiSdifferentfromalphabeticlettersSotheresearchonChinesecharacterrecognitiontechnologyisthekeytaskinthesocialinformatizationdevelopmentIntheinformationexplosionerahowtomakecomputerseffecti

3、vely“understand”SOmanyChinesecharacter，especiallyprintedChinesecharacterinordertosavemuchmanpower，isanimportantfactorinChinesecharacterrecognitiontechnologyImprovingtheeffectofprintedChinesecharacterrecognition，including

4、recognizingrateandrecognizingspeed，hasimportantpracticalandtheoreticalsignificanceinOA，machinetranslation，AIandSOonTakingthefirstclasstotally3755ChinesecharacterintheGB2312—80asthecharacterdatabase，thispaperelaboratesChi

5、nesecharacterrecognitiontechnologyfromfeatureextractionandmatchingrecognitionItdoescarefullyresearchonthesetwoaspectsandmakesimportantimprovementCompared、Ⅳitlloriginalalgorithms，theimprovedalgorithmsperformbetterThemaint

6、askofthepaperisasfollow(1)Ittakesstrengthsandweaknessesofbothstructurefeaturesandstatisticalfeaturesintoconsiderationandanalyzesmanykindsofcharacterfeatures，includingcomplexity，connectedcomponents，closedregions，coarseper

7、iphery，cellularfeature，texturefeatureandSOonBasedontheresearchofthefeatureextractionalgorithms，Itanalyzesproblemsandpresentscertainimprovedalgorithmsthatovercomeorweakentheproblemsandprovidescombinatorialoptimizationfeat

8、urewitIlapowerfulguaranteeMoreover，italsopresentsstrokescodefeatureandfeaturepointsbasedonChinesecharacterstrokestypes，andbothofthemimprovetherecognizingeffectsofthisChinesecharacterrecognitionsystem(2)Duringthematchingp

9、rocess，itelaboratesusualclusteringmethod，includingISODATA，UPGMA，KmeansandoneimprovedKmeansmethod，andpresentscertainresolutionsconsideringprimedChinesecharacterrecognitionFinally，thepaperpresentsanimprovedKmeansmethodbase