Recognitionsystem
Textrecognitiongenerallyincludesseveralpartssuchasthecollectionoftextinformation,theanalysisandprocessingofinformation,andtheclassificationofinformation.
InformationcollectionThegrayscaleofthetextonthepaperistransformedintoanelectricalsignal,whichisinputintothecomputer.Informationcollectionisrealizedbythepaperfeedingmechanismandphotoelectricconversiondeviceinthecharacterrecognitionmachine,suchasflyingspotscanning,camera,photosensitiveelement,laserscanningandotherphotoelectricconversiondevices.
Informationanalysisandprocessingeliminatesallkindsofnoiseandinterferencecausedbyprintingquality,paperquality(uniformity,stains,etc.)orwritingtoolsonthetransformedelectricalsignal,andperformssize,deflection,Variousnormalizationtreatmentssuchasshadeandthickness.
ClassificationanddiscriminationofinformationCarryoutclassificationanddiscriminationonthetextinformationafterthenoiseisremovedandnormalizedtooutputtherecognitionresult.
Recognitionmethod
TextrecognitionmethodThetextrecognitionmethodisbasicallydividedintothreecategories:statistics,logicaljudgmentandsyntax.Commonlyusedmethodsincludetemplatematchingandgeometricfeatureextraction.
①Thetemplatematchingmethodcorrelatestheinputtextwiththegivenstandardtext(template)ofeachcategory,calculatesthedegreeofsimilaritybetweentheinputtextandeachtemplate,andtakesthecategorywiththegreatestsimilarityasRecognitionresults.Thedisadvantageofthismethodisthatwhenthenumberofrecognizedcategoriesincreases,thenumberofstandardtexttemplatesalsoincreases.Ontheonehand,itwillincreasethestoragecapacityofthemachine,andontheotherhand,itwillalsoreducetheaccuracyofrecognition,sothismethodissuitableforrecognizingprintedcharacterswithfixedfonts.Theadvantageofthismethodistousetheentiretextforsimilaritycalculation,soithasstrongadaptabilitytotextdefectsandedgenoise.
②Thegeometricfeatureextractionmethodextractssomegeometricfeaturesofthetext,suchastheendpoints,bifurcationpoints,concaveandconvexpartsofthetext,horizontal,vertical,andinclinedlinesegments,closedloops,etc.,accordingtothesefeaturesThepositionandmutualrelationshipofthetwoarejudgedbylogicalcombination,andtherecognitionresultisobtained.Thisrecognitionmethodusesstructuralinformation,andisalsosuitableforcharacterswithlargedeformationssuchashandwrittencharacters.
Applicationfields
Textrecognitioncanbeappliedtomanyfields,suchasreading,translation,retrievalofdocuments,sortingoflettersandpackages,editingandproofreadingofmanuscripts,andalargenumberofstatisticalreportsAndthecollectionandanalysisofcards,theprocessingofbankchecks,thestatisticalsummaryofcommodityinvoices,theidentificationofcommoditycodes,themanagementofcommoditywarehouses,andthecollectionofwater,electricity,gas,rent,personalinsuranceandotherfees.Automaticprocessingandpartialautomationofofficetypists'work,etc.Aswellasdocumentretrievalandidentificationofvariousdocuments,itisconvenientforuserstoenterinformationquicklyandimprovetheworkefficiencyofallwalksoflife.
Commonlyusedsoftware
Textrecognition
ScanOCRtextrecognitionsoftware,supportall-roundscanning,photorecognitionandtranslationtechnology,isapicturetotextthatsupportstextextractionandPhotographliteracytranslationsoftwarewithtexteditingfunction.
Commonfunctions
1.Uploadpicturerecognition:Fasttextrecognitionsupportsdirectuploadofmobilephonephotoalbumpicturesandconvertthemintotext;
2,photorecognition
3,picturegenerationPDF
4,phototranslationp>
5.Ticketrecognition
6.Handwritingrecognition
CurrentsituationinChina
WiththecomprehensivedevelopmentofChina'sinformatizationconstruction,OCRtextrecognitiontechnologyhasbeenbornformorethan20years,andhasexperiencedthetransformationfromlaboratorytechnologytoproducts,andhasenteredthematurestageofindustryapplicationdevelopment.Comparedwiththewidespreadapplicationindevelopedcountries,theapplicationofOCRtextrecognitiontechnologyinallwalksoflifeinChinastillhasabroadspace.Asthenationalinformatizationconstructionentersthecontentconstructionstage,abrand-newindustryapplicationsituationhasbeencreatedforOCRtextrecognitiontechnology.LeadingChinesecharacterrecognitioncompaniessuchasWentong,YunmaiTechnologyandHanwangwillgodeeperintovariousfieldsofinformationconstruction.