基于FCN-AC-ASPP的手写体去除方法-豆柴文库

您所在位置：网站首页 / 基于FCN-AC-ASPP的手写体去除方法.docx / 文档详情

在线预览结束，喜欢就下载吧，查找使用更方便

5 金币

下载文档

/ 2

下载提示

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

文本预览

基于FCN-AC-ASPP的手写体去除方法
Title:HandwrittenCharacterRemovalbasedonFCN-AC-ASPP
Abstract:
Handwrittencharacterremovalisanessentialpreprocessingstepinmanydocumentanalysisandrecognitiontasks.ThispaperproposesamethodforremovinghandwrittencharactersfromtextdocumentsusingFCN-AC-ASPP(FullyConvolutionalNetworkwithAttentionCascadingandAtrousSpatialPyramidPooling).Theproposedmethodutilizesadeepconvolutionalneuralnetworktoaccuratelysegmentandremovehandwrittencharacterswhilepreservingtheunderlyingtextcontent.Experimentalresultsdemonstratetheeffectivenessofourmethodinachievinghigh-qualitycharacterremoval.
1.Introduction
Handwrittencharactersoftenappearindocumentimages,posingachallengetosubsequenttextrecognitiontasks.Theremovalofthesehandwrittencharactersisvitaltoimprovetheperformanceofopticalcharacterrecognition(OCR)systems.Traditionalmethodsforcharacterremovalrelyonrule-basedheuristicsandmanualinterventions,whicharetime-consuming,labor-intensive,andoftenresultinunsatisfactoryresults.Therefore,thereisaneedforanautomatedandaccuratemethodforremovinghandwrittencharactersfromtextdocuments.
2.LiteratureReview
Previousresearchoncharacterremovalhasmainlyfocusedonrule-basedortemplate-basedmethods,whichrelyonpriorknowledgeorhumaninterventions.Somemethodshaveutilizedtextureanalysisandmathematicalmorphologytosegmentandremovecharacters.However,thesemethodsoftenstrugglewithcomplexbackgrounds,varyingfontstyles,andoverlappingcharacters.Inrecentyears,deeplearning-basedapproacheshavegainedpopularityduetotheirabilitytolearndiscriminativefeaturesfromlarge-scaletrainingdata.FullyConvolutionalNetworks(FCNs)havebeensuccessfullyappliedtovariousvisualrecognitiontasks,includingimagesegmentation.
3.ProposedMethod:FCN-AC-ASPP
TheproposedmethodforhandwrittencharacterremovalisbasedonthecombinationofFCN,AttentionCascading,andAtrousSpatialPyramidPooling(ASPP).
3.1FCNArchitecture
TheFCNarchitectureisusedtoextracthigh-levelfeaturesfromtheinputdocumentimage.Itconsistsofmultipleconvolutionallayersfollowedbypoolinglaye