Seite - 50 - in Document Image Processing
Bild der Seite - 50 -
Text der Seite - 50 -
J. Imaging 2018,4, 57
in thepresentwork,wehaveamalgamatedtheconceptof ‘uniformpatterns’withRLBPtogenerate
RULBP.TheformaldefinitionofRULBPisgivenbelow:
RULBP(M,R)(xcen,ycen)= {
∑Mn=1 f(In− Icen− th), ifU(RILBP(M,R)(xcen,ycen))≥2,
M+1, otherwise. (10)
ThevalueofU(RLBP(M,R)(xcen,ycen)) is computedusingEquation(8).
2.5.2. SelectingtheValueof th
FromEquation(9), it canbe inferredthat the threshold(th) inRLBPplaysan important roleand
whosevaluemightbeapplicationspecific tosomeextent. Thus, in thiswork,wehaveattemptedto
rationalize it in thecontextof text/non-text separation inhandwrittendocuments.
Mosthandwrittendocumentsgenerallypossessa large intensityvariationat thestroke leveldue
tothevariednatureofwritinginstrumentsandnon-uniformityintheamountofpressureappliedwhile
writing. Thisnon-homogeneityoverasinglestrokecanonlybe identified ifwemagnify the image
(see thedarkandbrightpatcheswithin thestroke inFigure5. Forexample,LBPfor the3×3segment,
markedinred, inFigure5 is ‘00010001’.However, thevisualperceptionofahumanbeingconsiders
thisasahomogeneousregionwithall zeros ‘00000000’. Thispropertyofhandwrittendocumentsmay
generateerroneousLBPfeaturevalues,which, in turn, fail todistinguish the textcomponents fromthe
non-textones. Inorder tosolvesuchproblems,a threshold ‘th’hasbeen introducedinLBPtogenerate
RLBP.This thresholdensures that twograyvalues thatarenotperceptiblydifferentarenot labeled
differently. Theproblemwithselectingavalueof th is that, if thevalue isextremely large, thenthe
entire regionwillbehave likeahomogeneousregionwithnointensityvariation. This isbecause the
binarypattern ,according toEquation (10),will be all zeros for everypixel. Therefore,weneed to
provideanupper limit, thmax , on thevalueof th.
Figure5.Magnifiedimageofastrokeshowsthevariation ingrayvalues.A3×3matrixshowsthe
intensityvaluesof thegray imagesegmentmarkedinred.
Toaddress this issue,wehavesetanupper limit, thmax , onthevalueof th. Generally, inareal-life
handwrittendocument image, the intensityof thebackgroundpixels residewithinacloseproximity
of themaximumintensity255.Here,weassumethat the intensityof thebackgroundpixelswillbe ina
rangeof [245,255].Now,foreachimage,wefindthehighestgray-scale intensity(Igraymax) less than245.
Weclaimthat thepixelPhavingthis intensityvaluehas tobeapartof somewritingstroke. thmaxhas
tobesuchthat, ifweconsider Icenhasavalue Igraymax andaneighboringpixelhasavalue245, f(x)
as given inEquation (2) for x= In− Icen− thgives avalue 1. Therefore, thmax = 245− Igraymax.
Thevalueof thcanbeanythingbetween thmax and0.Wehaveperformedaweightedaverageof the
thresholdvalues in therange,with theweights increasingforhighervaluesof thandfoundthe ideal
50
zurück zum
Buch Document Image Processing"
Document Image Processing
- Titel
- Document Image Processing
- Autoren
- Ergina Kavallieratou
- Laurence Likforman-Sulem
- Herausgeber
- MDPI
- Ort
- Basel
- Datum
- 2018
- Sprache
- deutsch
- Lizenz
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03897-106-1
- Abmessungen
- 17.0 x 24.4 cm
- Seiten
- 216
- Schlagwörter
- document image processing, preprocessing, binarizationl, text-line segmentation, handwriting recognition, indic/arabic/asian script, OCR, Video OCR, word spotting, retrieval, document datasets, performance evaluation, document annotation tools
- Kategorie
- Informatik