基于HMM的脱机手写大写金额识别中的单分类器性能分析.pdfVIP

2006中国控制与决策学术年会论文集 Decision 343 Chinese。“耐andCon．fereace of2006 Proceedingj 基于HMM的脱机手写大写金额识别中的单分类器性能分析 王先梅1，林子钰2 (1．北京科技大学电子信息工程系，北京1000B$；2．北京中环冶金总公司。北京100011) 摘要；以无约束脱机手写大写叠瓤识别为研兜对皇．构建喜种基于HMM曲单分类喜．为了考察不同特征对分类 的影响。每十单分类墨均使用一种单矬的特征．通过对茸行收集曲样本库进行测试t得到了各种单分类毒对脱机手写 大写盎颧的识别性能，并进行了比较．实验结果袁螗，笔划密度特征、粗外围特征以及方】旬线紊特征的识烈性能较住t 草分类孽的识羽率均在93％以上，而且对各十模盘类曲识别率也较高． 关键词；脱机手写大写叠颧识别#HMM；分类毒}特征提取 Performanceof Classifierin of AnalysisSingle Recognition inWordsBasedonHMM Off—lineHandwrittenAmount WANG X如”me／1。UⅣ五一yu2 ofScienceand 100083· ofElectronicInformation Technology，Beijing (1．Department Engineering，University China，2．Zhon‘ghuaoMetallurgical corn) plum-wang@ton based arebuilt oItekindoffeathersfor of AbstractlSome classifierson}玎证M byusing recognition ingle toobservethe ofeachkindoffentures· unconstrainedoff-linehandwrittenamountinwords．Inorder performance showthatstroke featuret classifier odekindoffeatures．Theexperimentalresults density every tJse$jlist single for off-linehandwrittenamountin featureanddirectionallineelementfeaturearesuitable of peripheral recognition words．Each rateofthesethreeclassifierSisover93％． recognition off-linehandwrittenamountin classifier；Featureextraction words：Unc