Seite - 31 - in Document Image Processing

J. Imaging 2018,4, 80 homogenousslant. Itdoesnot requirepagesegmentation into text linesorwords. Thismakes the proposedtechniqueappropriate forhistoricaldocuments, especially formalones, since theyensure auniformslantover theentirepage.Moreover, thesegmentation into text lineswouldcreatemore noise. Usually, formal historical documents arewrittenbywell-educatedpeoplewith a standard writing style andfixed slant. On the other hand, the proposedmethodology is inappropriate for document imagescontainingunconstrainedwritingandseveral slantangles.Methodologies, such as theonedescribedin[6], aremoreappropriate insuchcases. Thus,experimentsareperformedon severaldatabases: • theTrigraphSlantdatabase [18] (theonlyavailabledatabase forslantestimation), • two databases of historical documents (George Washington [19] and Barcelona historical, handwrittenmarriagesdatabaseBH2M[20]) • asyntheticprinteddatabasewhereslantsare fullydetermined. Thecontributionof thecurrentworkconsistsof: 1. To thebestofourknowledge, this is thefirst time thata slant removal technique isproposed, able tobeappliedto theentirepage,withoutrequiringtext lineorwordsegmentation. 2. It does not generate extra noise, due to line and/orword segmentation thatwould remain in thepage after slant removal,which is accomplishedby shifting the entirepageuniformly andensuring texthomogeneity. Mostof theexisted techniquesapply to idealdatabases, like IAM-DB(Figure1) that is appropriatelymade for lineandwordsegmentation. In thecaseof historical documents (Figure 2), thefinal resultwouldbe full of dots and strokes because of thesegmentation. 3. Instructionsaregivenover thebestapplicationtodocumentpage,afterdetailedresults. In Section2.1, a shortdescriptionof the elaborated slantdetectionalgorithm[9] is presented. Theproposedtechnique isdescribed indetail inSection2.2,where theparameters thatareexamined indetail areanalyzed. Theexperimental results arepresentedandanalyzed inSection3while the conclusionsarediscussed inSection4. Figure2.Adocument fromtheBarcelonahistorical,handwrittenmarriagesdatabase (BH2M)[20]. 31