RetrievingOCRText: ASurveyofCurrentApproaches - - PowerPoint PPT Presentation

RetrievingOCRText: ASurveyofCurrentApproaches InformationRetrievalLab IllinoisInstituteofTechnology

• S. Beitzel

E.Jensen D.Grossman {steve,ej,grossman}@ir.iit.edu

2

Overview

• ModelsforOCRText
• ProcessingOCRTextforCategorization
• Auto-correctionofOCRErrors

3

ModelsforOCRText

• Mittendorf,Schauble,andSheridan(1995,1996)
• IncorporateprobabilitiesoftypicalOCRerrors
• Harding,Croft,Weir(1997)

– Additionofcharacter-basedn-gramstothemodel. – Ex:Environment

• _enenvnviviriroonmnmemenent– 3-grams

4

Auto-CorrectionofOCRErrors

• Liu(1991)

– Classifyeachtypeoferror – Usedictionarylookuptoidentifycandidate terms

• Taghva,Borsack andCondit(1994)

– Clusteringtogroupmis-spellingsinwiththeir correctlymis-spelledterms

5

OCRTextforCategorization

• Hoch (1994)

– UseofcategorizeronOCRtext,showed degradedperformancewithOCRdata.

• JunkerandHoch (1997)

– N-gramswereusedtoshowsomeimprovement aswellin[Junk97].

6

Summary

• ModelsexistforOCRretrieval
• N-gramshavebeenshowntohavesome

success

• NolargestandardtestcollectionofOCR

data,smallcollectionsexistwithsomeearly TRECdata.