Page - 22 - in Document Image Processing

J. Imaging 2018,4, 27 ThenormalizationconstantKisgivenas: K= 1 ∑x+di=x−d∑ x+d j=y−dGs(i;x, j;y)Gr[I(i, j), I(x,y)] (6) Equations (4)and(5) showthat thebilateralfilterhas threeparameters: σ2s (thefilterdomain), σ2r (thefilter range),andthe thirdparameter is thewindowsizeN×N[15]. Thegeometric spreadof thebilateral filter is controlledbyσ2s . If thevalueofσ2s is increased, more neighbours are combined in the diffusion process yielding a “smoother” image, while σ2r represents thephotometric spreading. Onlypixelswithapercentagedifferenceof less thanσ2r are processed[13]. 2.4.OtsuFiltering Afterpassing throughthebilateralfilter, the image is split into itsoriginal (non-gammacorrected) Red,GreenandBluecomponents,asshownintheblockdiagraminFigure2.Thekernelof thebilateral filter alters the balance of the colors in the original image in such away towiden thedifferences betweenthecolorof thefrontandback-to-front interference.AmodifiedversionofOtsu[16]algorithm is applied to each RGB channel using the thresholds determined by theDecisionMaking Block, whichmaybeconsideredas the“optimal” threshold for eachRGBchannel, and then threebinary imagesaregenerated. 2.5. ImageClassification Theimageclassificationblockwasalsotrainedwiththesynthetic images insuchawaytoanalyze the threebinary imagesgenerated ineachof thechannelsandoutputs theonethat isconsideredthe bestone. ThisdecisionwasalsomadebyanaïveBayesautomaticclassifierwhichwastrainedusing thecalculatedco-occurrencematrix foreachof the32,000synthetic imagesbycomparingeachof them with theoriginalgroundtruth image, theFront image. 3. ExperimentsandResults As already explained, the enormous variety of kinds of text documents makes extremely improbable that one single algorithm is able to satisfactorily binarize all kinds of documents. Depending on the nature (or degree of complexity) of the image several or no algorithm will be able to provide good results. This paper follows the assessment methodology proposed in reference [9], inwhichone compares thenumbers of backgroundand foregroundpixels correctly matchedwith aground-truth image. Twenty-three binarization algorithmswere testedusing the methodologydescribed: 1. Mello-Lins [5] 2. DaSilva-Lins-Rocha[6] 3. Otsu[16] 4. Johannsen-Bille [17] 5. Kapur-Sahoo-Wong[18] 6. RenyEntropy(variationof [18]) 7. Li-Tam[19] 8. Mean[20] 9. MinError [21] 10. Mixture-Modeling[22] 11. Moments [23] 12. IsoData [24] 13. Percentile [25] 22