Seite - 33 - in Document Image Processing
Bild der Seite - 33 -
Text der Seite - 33 -
J. Imaging 2018,4, 80
Inmoredetail, thedetectionalgorithm[9]consistsof thesteps:
1. Thewordimageisartificiallyslantedtoboth, leftandright,underdifferentslantdetectionangles.
Themaximumslantangle isapproximately45degreesandtheslantanglestepdependsonthe
heightof the text image.
2. Foreachof theextractedwordimages, theverticalprojectionprofile iscalculated.
3. TheWVDiscalculatedforall theaboveprojectedprofiles.
4. Thecurvesofmaximumintensityof theWVDsareextracted, justbykeepingthemaximumvalue
ofeachcurveof thespace-frequencydistribution, for thespecificslant.
5. Thecurveofmaximumintensitywith thegreatestpeak,correspondingto theprojectedprofile
with themost intensealternations is selected.
6. Thecorrespondingwordimage isselectedas themostnon-slantedword.
Theaboveprocedure is repeatedtwice,once forabigstepsizeof10degrees (BigStep)where the
areaaroundanANGLE1isselectedcloser to theslantandthesecondtimeforasmaller stepsizeof1
degree,whereamoredetaileddetection isperformedandamoreexactareaANGLE2isdetected.
Thisway, the computational cost is reduced, since the first detection is performed between
fewerpossibleangles inorder to localizeroughly theareaANGLE1,beforeamoreaccuratedetection
(ANGLE2) isperformed in this specific area for a step sizeof onedegree. Aslant of less thanone
degree isnotconsidered importantenoughtobeexamined. Finally, thedetectedangle (Detected_Slant)
isgivenby
Detected_Slant=ANGLE1×BigStep+ANGLE2×1 (2)
2.2. ProposedSlantRemovalTechnique
Theproposedtechnique isbasedontheslantdetectionalgorithmpresentedinSection2.1,but
inourcase, it isapplied to text fragments insteadofwords (Figure4). It isbasedonthe fact that in
historical documents there is auniformslant that extends throughout the entiredocument image.
Sincenosegmentation isperformed, fragmentsof textareusedinsteadofwords.
/RFDOL]DWLRQ RI
IUDJPHQWV +[:
6ODQW
'HWHFWLRQ
'RFXPHQW 6ODQW
5HPRYDO
'RFXPHQW
,PDJH
'HVODQWHG
7H[W ,PDJH
Figure4.Theproposedslant removal techniqueappliedto fragmentsof textcorrects theentirepage
withoutsegmentation.
33
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