Seite - 168 - in Document Image Processing

J. Imaging 2018,4, 39 5.Conclusions Thisisthefirstapplicationofclassifiercombinationapproachesinthedomainofscriptrecognition consideringthenumberofscriptsbeingundertakenandtherangeofclassifiercombinationprocedures that are evaluated. Combination is performed at the feature level aswell as decision level using abstract level, rank levelandmeasurement level informationprovidedbytheclassifiers. Encouraging resultsareobtainedfromtheexperiments.Highaccuracies in therangeof95–98%havebeenachieved byusing combination techniques as shown in thepreviousResult section. There is an increaseof over7%withthebestperformingMLPclassifierwhenLogisticRegression isusedas thesecondary classifier for7200samples from12different scripts. So, thismodelproves tobeuseful for thiscomplex patternrecognitionproblemandmakesabetterdecisionbasedonthe informationprovidedbythe baseclassifier. Though, in thepresentwork, threesourcesof informationwithdifferent featuresetshavebeen combinedusing their respective classifier resultsbut thisprocess canbeextended to includemore inputsourcesalongwithdifferentclassifier.With the increase in thenumberofsources,an intelligent anddynamicselectionprocedureneeds tobeemployedinorder to facilitatecombination inamore meaningfulway. The combination being an overhead to the classification task, it is important to developmethods that can indicate if the combinationwouldworkor not qualitatively. In future, theworkcanbeextendedforalargerdatasetsothattherobustnessoftheprocedurescanbeestablished. Thescriptrecognitionsystemhereisageneral frameworkwhichcanbeappliedtoothersimilarpattern recognition tasks likeblockandline level recognitionofscripts toestablish itsusefulness indocument analysis research. Acknowledgments:Theauthorsarethankful totheCenter forMicroprocessorApplicationforTrainingEducation andResearch(CMATER) andProjectonStorageRetrievalandUnderstandingofVideoforMultimedia (SRUVM) ofComputerScienceandEngineeringDepartment, JadavpurUniversity, forproviding infrastructure facilities duringprogressof thework. Theauthorsof thispaperarealso thankful toall those individualswhowillingly contributed indevelopingthehandwritten Indicscriptdatabaseusedin thecurrent research. Author Contributions: Anirban Mukhopadhyay and Pawan Kumar Singh conceived and designed the experiments; Anirban Mukhopadhyay performed the experiments; Anirban Mukhopadhyay and PawanKumarSinghanalyzedthedata;RamSarkaramdMitaNasipuri contributedreagents/materials/analysis tools;AnirbanMukhopadhyayandPawanKumarSinghwrote thepaper. Conflictsof Interest:Theauthorsdeclarenoconflictof interest. The foundingsponsorshadnorole in thedesign of the study; in the collection, analyses, or interpretationofdata; in thewritingof themanuscript and in the decisiontopublish theresults. References 1. Singh,P.K.;Sarkar,R.;Nasipuri,M.OfflineScript IdentificationfromMultilingual Indic-scriptDocuments: Astate-of-the-art.Comput. Sci. Rev. 2015,15–16, 1–28. [CrossRef] 2. Ubul,K.; Tursun,G.;Aysa,A.; Impedovo,D.; Pirlo,G.;Yibulayin,T. Script IdentificationofMulti-Script Documents:ASurvey. IEEEAccess2017,5, 6546–6559. [CrossRef] 3. Spitz,A.L.Determinationof thescriptandlanguagecontentofdocument images. IEEETrans. PatternAnal. Mach. Intell. 1997,19, 234–245. [CrossRef] 4. Tan,T.N.RotationInvariantTextureFeaturesandtheiruse inAutomaticScript Identification. IEEETran. PatternAnal.Mach. Intell. 1998,20, 751–756. [CrossRef] 5. Hochberg, J.;Kelly,P.;Thomas,T.;Kerns,L.Automaticscript identificationfromdocument imagesusing cluster-basedtemplates. IEEETrans. PatternAnal.Mach. Intell. 1997,19, 176–181. [CrossRef] 6. Hochberg, J.;Bowers,K.;Cannon,M.;Keely,P.Scriptandlanguageidentificationforhand-writtendocument images. IJDAR1999,2, 45–52. [CrossRef] 7. Wood, S.; Yao, X.; Krishnamurthi, K.; Dang, L. Language identification for printed text independent of segmentation. InProceedingsof the InternationalConferenceonImageProcessing,Washington,DC,USA, 23–26October1995;pp.428–431. 168