資源描述:
《仿射傳播聚類自適應手寫字符識別》由會員上傳分享,免費在線閱讀,更多相關內容在行業(yè)資料-天天文庫。
1、JournalofComputerApplicationsISSN1001-90812015-03-10計算機應用,2015,35(3):807-810CODENJYIIDUhttp://www.joca.cn文章編號:1001-9081(2015)03-807-04doi:10.11772/j.issn.1001-9081.2015.03.807基于仿射傳播聚類的自適應手寫字符識別*楊怡,王江晴,朱宗曉(中南民族大學計算機科學學院,武漢430074)(*通信作者電子郵箱wjqing2000@aliyun.com)摘要:對于手寫字符識別過程中相似字符較多且相同字符存在大量不
2、規(guī)則書寫變形的問題,提出一種改進的仿射傳播聚類算法加入手寫字符識別過程中。該算法基于原始仿射傳播(AP)聚類算法,將其與聚類評判函數(shù)Silhouette結合,通過AP算法迭代過程自適應地改變偏向參數(shù)以調整類別數(shù),并且結合每次聚類質量得到最優(yōu)聚類結果?;谑謱憹h字識別的實驗結果表明,加入了原始AP算法的識別率比傳統(tǒng)識別過程得到的識別率總體提高1.52%,而加入改進AP算法的識別率又比加入原始AP算法的識別率總體提高了1.28%。該實驗結果驗證了加入聚類算法于手寫字符識別過程的有效性,而改進AP算法相比原始AP算法在收斂性和聚類質量上都有一定的提高。關鍵詞:仿射傳播聚類;手寫字
3、符;評判函數(shù);偏向參數(shù);聚類質量中圖分類號:TP391.1文獻標志碼:AAdaptivehandwrittencharacterrecognitionbasedonaffinitypropagationclustering*YANGYi,WANGJiangqing,ZHUZongxiao(CollegeofComputerScience,South-CentralUniversityforNationalities,WuhanHubei430074,China)Abstract:Fortoomanysimilarwordsandlotsofirregularwritingw
4、aysofthesamewordsinthehandwrittencharacterrecognition,amodifiedAffinityPropagation(AP)clusteringalgorithmwasproposedtoaddtotherecognitionprocess.ClusteringjudgingfunctionSilhouettewascombinedwithoriginalAPalgorithmintheproposedalgorithm.Classnumberwasupdatedbychangingpreferenceparameterada
5、ptivelythroughiterativeprocessofAPalgorithm.Andthentheoptimalclusteringresultwasobtainedbyassessingclusteringqualityofeveryiteration.TheexperimentofhandwrittenChinesecharacterrecognitionindicatesthattherecognitionrateofrecognitionprocessaddedoriginalAPalgorithmis1.52%higherthantherateoftra
6、ditionalrecognitionprocess.AndtherecognitionrateofrecognitionprocessaddedmodifiedAPalgorithmis1.28%higherthantherateofrecognitionprocessaddedoriginalAPalgorithm.Theexperimentalresultsverifythatitiseffectivetoaddclusteringalgorithmtothehandwrittencharacterrecognitionprocess.Andcomparedwitho
7、riginalAPalgorithm,convergenceandclusteringqualityofmodifiedAPalgorithmarealsoimproved.Keywords:AffinityPropagation(AP)clustering;handwrittencharacter;judgingfunction;preferenceparameter;clusteringquality漢字識別是人工智能和模式識別中重要的組成部分,脫機手寫漢字識別是當前模式識別研究領域中最具挑戰(zhàn)性的任務之一。由于