Page - 178 - in Document Image Processing
Image of the Page - 178 -
Text of the Page - 178 -
J. Imaging 2017,3, 62
3.5.AdaptiveBlur
Theblurdefect isaverycommondefectencounteredduringdigitizationcampaigns. Thedifficulty
here is tocreatearealisticblurdefect thatmimics theveryslightblur thatappearswhenthescanner is
incorrectlyset (a largeblur iseasilydetectedbyscanners). Todoso,weproposeamethodinspiredby
theblurdetectionfrom[50]. First, theuserchoosesablur imagetomimicamongarealblurexample
available inDocCreator. Then,usingadichotomicalgorithm,wecompute thesizeof thekernelof
aGaussianblur that,onceappliedonthe input image,producesablursimilar to thechosenrealblur
image. In thismethod, theFourierTransformof the image isfirst computed. Thenthemoduleof the
FourierTransformisbinarizedaccordingto itsmean. Theresultingbinarized imageproducesadisc
for imageswithonly text.As thehighfrequenciesdecreasewhentheblur increases, thedisc radius in
thebinarized imagealsodecreaseswhen theblur increases. This radius isused tocharacterize the
images. Thedichotomicalgorithmisusedtosearchthekernel size thatproducesaradiussimilar to
theonefoundontheselectedexample image. SeeFigure7 foranexample.
Figure 7. Adaptiveblurdefect. (Left) imagewith real blur. (Right) imagewith synthetic blur that
mimics therealone
3.6. 3DPaperDeformation
Thepaperonwhichabookisprintedmayhaveseveral typesofdeformation(alongcurvature,
rotation, fold,hole, etc.).Weproposea3Ddeformationmodel thatgenerate realistic smallor large
paperdeformations.
Thefullprocessisdetailedin[51]. Themainideais: first,a3Dscannerisusedtoacquirea3Dmesh
fromarealdocument. Thismeshpreservesall representativedistortions. Then, themesh isunfolded
intoa2Dplan. Therefore,eachvertex in themeshhasacorresponding2Dpoint. Thecoordinatesof
suchapoint are consideredas texture coordinates. Finally, themeshcanbe renderedwithany2D
imagemappedasa texture. For therendering,weuse thePhongreflectionmodelas the illumination
model. Changing lightproperties andpositionallows toaccentuateorminimizedistortioneffects.
DocCreatorcurrentlyprovides17parameterizedmeshes, enablingone toproducenumerousdistorted
images. Figure8showssuchadeformation.
This3Dpaperdeformationmodelcanbeusedtosimulatemobiledocumentcapture. Theusercan
addabackgroundplanewith texture,ontopofwhichthedocumentstands. Bychangingviewpoint
andlightpositions, theusercangeneratemanyimages. These imagescanbeusedforcamera-based
document imageanalysis. Figure9showsexamplesof twopointsofviewgeneratedwith thesame
document image.
178
back to the
book Document Image Processing"
Document Image Processing
- Title
- Document Image Processing
- Authors
- Ergina Kavallieratou
- Laurence Likforman-Sulem
- Editor
- MDPI
- Location
- Basel
- Date
- 2018
- Language
- German
- License
- CC BY-NC-ND 4.0
- ISBN
- 978-3-03897-106-1
- Size
- 17.0 x 24.4 cm
- Pages
- 216
- Keywords
- 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
- Category
- Informatik