OCR OF HANDWRITTEN DEVANAGARI TEXT USING SEGMENTATION AND CNN

Abstract

Nepali LanguageusesDevanagariscriptwhichisbeingusedbeyondparticularlanguageas officialscriptinIndiansubcontinent.CharacterclassificationaccuracyisstillanissuetoDe- vanagariScriptdespiteofmanyresearcheshadbeenconducted.Throughmanyresearches conducted, itisidentifiedthatpoorsegmentationisthekeytolowaccuracy.Manyresearch worksusedholisticapproachtoimproveclassificationresult. DevanagariisthemostextensivelyusedscriptinNepal.Textdetectionandextractionfromthe images/photos havebeenutilizedinavarietyofapplications.Characterrecognitionpossessthe greater levelofpotentialtoleadtotechnologicallyadvancedapplications.However,duetothe structure’scomplexity,thesametechniquecannotbeusedforalllanguagesandscripts.The OCR methodstransformsthescannedimagesofprintedorhandwrittendocumentintothedig- ital text.Languages,includingNepali,Hindi,Marathi,andothers,usetheDevanagariscript. The majorgoaloftheproposedwork/studyistodetectNepali(Devanagari)scriptusingCNN utilizing themostappropriatestrategiestoboosttherecognitionrateofDevanagariHandwrit- ten TextRecognitionwithaviewtoeffectivelyrecognizeDevanagariTextfromthescanned images. OCRbeginswiththeinputandpreprocessingofscannedpicturesofprintedorhand- written text.Classifiersareusedtocategorizethepreprocessedpicture,whichisseparatedinto wordsandthencharactersbyutilizingprofilingprojectiontechnique.Finally,post-processing involvestheprocessoftransformingdetectedcharactersintotext. This projectsalsorepresentsthemethodofskew-correctionforcorrectingtheskewofthewords for theactualprediction.OCRmodelpresentedinthisresearchprojectusesprofileprojection (Horizontal andVertical)ofthepurposeofsegmentationofcore,lowermodifiersandupper modifiers.