Page - 184 - in Document Image Processing

J. Imaging 2017,3, 62 detailed in this paper showsemi-synthetic and synthetic documents createdwithDocCreator are useful forperformanceevaluation, retrainingtasksorperformanceprediction. In futurework,weplan to improve thesyntheticdocumentcreation toavoid tohave toodifferentcharacters in thecomposed document. Forexample,weshould investigate if addingsomeconstraintsonthe fontextractionphase or taking intoaccount thecontextwhenaddingnewcharacters to thesyntheticdocumentmayleadto morerealistic syntheticdocuments.Wealsoconsider tosetupacognitiveexperiment toevaluate the perceivedrealnessof thedegradeddocumentsoreventhecreatedsyntheticdocuments.Wearealso planningto investigatehowthegenerationofhighlydiversifieddatacan improvetheresultsof tasks basedondeeplearningmethods. DocCreator (source,Linux,Mac,Windowspackagedversions),all thedatabasesusedfor thetests, avideoandanextradatabase (31.000synthetic imagesgeneratedwithWilliamShakespearesonnet textfiles)areavailableat [http://doc-creator.labri.fr/]. AuthorContributions:Nicholas Journet,MurielVisaniandBorisMansencalcontributed inequalproportionto thecreationandtestsofDocCreator (modeldegradationandsyntheticdocumentreconstruction) ; theyalsowrote thisarticle.KieuVan-Cuongcontributedto thedegradationmodels,AntoineBillycontributedto thesynthetic documentreconstruction. Conflictsof Interest:Theauthorsdeclarenoconflictof interest. References 1. L’AffaireAlexis. Availableonline: http://gallica.bnf.fr/ark:/12148/bpt6k8630878m/f1.item.texteImage (accessedon9December2017). 2. Shahab,A.;Shafait,F.;Kieninger,T.;Dengel,A. AnOpenApproachTowards theBenchmarkingofTable StructureRecognition Systems. InProceedings of the 9th IAPR InternationalWorkshoponDocument AnalysisSystems,Boston,MA,USA,9–11 June2010;ACM:NewYork,NY,USA,2010;pp. 113–120. 3. Lazzara, G.; Levillain, R.; Géraud, T.; Jacquelet, Y.; Marquegnies, J.; Crépin-Leblond, A. The SCRIBO Moduleof theOlenaPlatform: aFreeSoftwareFrameworkforDocument ImageAnalysis. InProceedings of the 2011 International Conference onDocumentAnalysis andRecognition (ICDAR), Beijing, China, 18–21September2011. 4. Yalniz, I.;Manmatha,R. AFastAlignmentSchemeforAutomaticOCREvaluationofBooks. InProceedings of the 2011 International Conference onDocumentAnalysis andRecognition (ICDAR), Beijing, China, 18–21September2011;pp. 754–758. 5. Roy,P.;Ramel,J.;Ragot,N.WordRetrievalinHistoricalDocumentUsingCharacter-Primitives. InProceedings of the 2011 International Conference onDocumentAnalysis and Recognition (ICDAR), Beijing, China, 18–21September2011;pp. 678–682. 6. IAMHandwritingDatabase.Availableonline: http://www.iam.unibe.ch/fki/databases/iam-handwriting- database (accessedon9December2017). 7. Grosicki, E.;Carré,M.;Brodin, J.M.;Geoffrois,E. Resultsof the secondRIMESevaluationcampaign for handwrittenmail processing. In Proceedings of the 2009 10th International Conference onDocument AnalysisandRecognition(ICDAR),Barcelona,Spain,26–29 July2009. 8. Perez, D.; Tarazon, L.; Serrano, N.; Castro, F.; Terrades, O.R.; Juan, A. The GERMANA Database. InProceedingsof the200910thInternationalConferenceonDocumentAnalysisandRecognition(ICDAR), Barcelona,Spain,26–29 July2009; IEEEComputerSociety:Washington,DC,USA,2009;pp. 301–305. 9. Nakagawa,K.;Fujiyoshi,A.;Suzuki,M.Ground-truthedDatasetofChemicalStructure Images in Japanese PublishedPatentApplications. In Proceedings of the 9th IAPR InternationalWorkshoponDocument AnalysisSystems,Boston,MA,USA,9–11 June2010;ACM:NewYork,NY,USA,2010;pp. 455–462. 10. Eurecom.Availableonline: http://www.eurecom.fr/huet/work.html (accessedon9December2017). 11. UniversityofCalifornia,SanFrancisco. TheLegacyTobaccoDocumentLibrary (LTDL);UniversityofCalifornia: SanFrancisco,CA,USA,2007. 12. Delalandre,M.;Valveny,E.;Pridmore,T.;Karatzas,D. GenerationofSyntheticDocuments forPerformance EvaluationofSymbolRecognition&SpottingSystems. Int. J.Doc.Anal. Recognit. 2010,13, 187–207. 184