Page - 158 - in Document Image Processing

J. Imaging 2018,4, 39 3.1.4.Dempster-ShaferTheoryofEvidence TheDSframework[39] isbasedontheviewwherebypropositionsarerepresentedassubsetsof agivensetW, referredtoasa frameofdiscernment. Evidencecanbeassociatedtoeachproposition (subset) toexpress theuncertainty (belief) thathasbeenobservedordiscerned. Evidence isusually computed based on a density function m called Basic Probability Assignment (BPA) and m(p) represents thebeliefexactlycommittedto thepropositionp. DS theoryhasanoperationcalledDempster’s rule of combination that aggregates two (ormore) bodiesofevidencedefinedwithin thesameframeofdiscernment intoonebodyofevidence. Letm1 andm2 be twoBPAsdefinedinW. Thenewbodyofevidence isdefinedbytheBPAm1,2 as: m1,2(A)= { 0 if A=∅ 1 1−K ∑ B∩C=A m1(B)m2(C) if A =∅ (3) where,K=∑B∩C=∅m1(B)m2(C)andA is the intersectionofsubsetsBandC. Inotherwords, theDempster’s combinationrulecomputesameasureofagreementbetweentwo bodiesofevidenceconcerningvariouspropositionsdeterminedfromacommonframeofdiscernment. Therule focusesonlyonthosepropositions thatbothbodiesofevidencesupport. The denominator is a normalization factor that ensures that m is a BPA, called the conflict. TheYagar’smodificationoftheDStheory[40]hasbeenimplementedinthepaperwiththenormalizing factoras1. This reducessomeof the issuesregardingtheconflict factor. Earlier, DS theory based combination has been applied on different fields like handwritten digit recognition [41], skin detection [42], 3D palm print recognition [43] among other pattern recognitiondomains. 3.2. SecondaryClassifierBasedCombinationTechniques Theconfidencevaluesprovidedbytheclassifiersactas the featureset for thesecondaryclassifier whichactsonthesecondstageof the framework.With the trainingfromtheclassifierscores, it learns topredict theoutcomeforasetofnewconfidencescoresfromthesamesetofclassifiers. Theadvantage ofusingsuchagenericcombinator is that itcanlearnthecombinationalgorithmandcanautomatically account for the strengths and score ranges of the individual classifiers. For example,Dar-Shyang Lee[29]usedaneuralnetworktooperateontheoutputsof the individualclassifiersandtoproduce thecombinedmatchingscore.Apart fromtheneuralnetwork,otherclassifiers likek-NN,SVMand RandomForesthavebeenfittedandtested in thispaper. 4.ResultsandInterpretation 4.1. PreparationofDatabase Atpresent,nostandardbenchmarkdatabaseofhandwritten Indic scripts is freelyavailable in the publicdomain.Hence,wehavecreatedourowndatabaseofhandwrittendocuments in the laboratory. Thedocumentpages for thedatabasewerecollectedfromdifferentsourcesonrequest. Participants of thisdata collectiondrivewereasked towrite few linesonA-4 sizepages. Noother restrictions were imposedregardingthecontentof the textualmaterials. Thedocumentswerewritten in12official scriptsof India. Thedocumentpagesaredigitizedat300dpiresolutionandstoredasgreytone images. Thescanned imagesmaycontainnoisypixelswhichare removedbyapplyingGaussianfilter [33]. The textwordsareautomaticallyextractedfromthehandwrittendocumentsbyusingapage-to-word segmentationalgorithmdescribedin[44].Asamplesnapshotofwordimageswritten in12different scripts isshowninFigure7. Finally,atotalof7200handwrittenwordimagesareprepared,withexactly 600 textwordsperscript. 158