预览加载中,请您耐心等待几秒...
1/3
2/3
3/3

在线预览结束,喜欢就下载吧,查找使用更方便

如果您无法下载资料,请参考说明:

1、部分资料下载需要金币,请确保您的账户上有足够的金币

2、已购买过的文档,再次下载不重复扣费

3、资料包下载后请先用软件解压,在使用对应软件打开

基于LSTM型RNN的CAPTCHA识别方法 Abstract: CAPTCHA(CompletelyAutomatedPublicTuringtesttotellComputersandHumansApart)isintroducedtopreventautomatedtoolsfromperformingunauthorizedactionsonawebpage.Inthispaper,weproposeanewmethodforCAPTCHArecognitionbasedonLongShort-TermMemory(LSTM)typeRecurrentNeuralNetworks(RNNs).Theproposedmethodusesapre-processingsteptopreparetheCAPTCHAimagesbyconvertingthemintobinaryformandremovingnoise.Then,thesepre-processedimagesarefedintoaLSTM-basedRNNthatlearnstorecognizetheCAPTCHAs.TheresultsdemonstratethatourproposedmethodoutperformsotherCAPTCHArecognitionmethodsintermsofaccuracyandefficiency. Keywords:CAPTCHA,LSTM,RNN,recognition,pre-processing. Introduction: CAPTCHAisapopulartechniqueusedtopreventautomatedtoolsfromperformingunauthorizedactionsonawebpage.CAPTCHAstandsfor“CompletelyAutomatedPublicTuringtesttotellComputersandHumansApart”.CAPTCHAinvolvesdisplayingdistortedimagesofalphanumericcharactersthatahumancaneasilyrecognizebutaredifficultformachinerecognition.CAPTCHAsareusedtopreventspamming,bruteforceattacks,andothermaliciousactivitiesonwebsites. CAPTCHAisachallengingtaskforautomatedrecognitionsystemsduetoitsnoisyanddistortednature.Therefore,developingarobustCAPTCHArecognitionsystemhasbecomeanimportantareaofresearchincomputervisionandmachinelearning.Inthispaper,weproposeanewmethodforCAPTCHArecognitionbasedonLongShort-TermMemory(LSTM)typeRecurrentNeuralNetworks(RNNs).Theproposedmethodusesapre-processingsteptopreparetheCAPTCHAimagesbyconvertingthemintobinaryformandremovingnoise.Then,thesepre-processedimagesarefedintoanLSTM-basedRNNthatlearnstorecognizetheCAPTCHAs. Relatedwork: SeveraltechniqueshavebeenproposedforCAPTCHArecognitionintheliterature.Someofthesetechniquesincludeimageprocessingtechniques,patternrecognitiontechniques,neuralnetwork-basedtechniques,anddeeplearning-basedtechniques. Imageprocessingtechniquesinvolveimagefiltering,imagemorphologicaloperations,andothertechniquestoremovenoiseandenhancetheimages.Patternrecognitiontechniquesinvolv