Page - 34 - in Document Image Processing
Image of the Page - 34 -
Text of the Page - 34 -
J. Imaging 2018,4, 80
Aspreviouslymentioned, in thepast, educatedpersons tookspecial carewhenwriting, resulting
in ahighdegreeof stability in the slant of theirwriting style. Thus, inorder todetect the slant of
the text inahistoricaldocumentpage,a fewfragmentsof textareconsidered.Althoughonesample
couldtheoreticallybeenough, severalonesare inpracticenecessary toensurecoverageofpageswith
sparse textorspecial formatting, suchascolumns,arrays,etc. To localizeappropriate fragments the
followingwayis followed: apage isscannedfromleft toright, toptobottomusingawindowofsize
HxW (heightXwidth), startingfromthepixelposition(skip, skip) inorder toskipscanningorother
noise. Skipcanbegeneral e.g., 1/5ofdocumentwidth (here), orbedetermineddependingon the
collection.Allblackpixels (black_pixels) inside thewindowarecounted. Thearea inside thewindowis
retained ifCondition (3) is trueandthescanningstopswhenthe requirednumberMof fragments
is localized.
black_pixels
HxW >R (3)
TheCondition (3) requires the text in thewindow to takeupmore thanR=0.10 of the area.
Thesizeof thewindowintheseexperiments,HxW,wasselectedasH=2mbandW=7mb,where
mbis themaincharacterbodysize in thepage(heightof thecharacterbodyexcludingascendersand
descenders). In thecurrentpaper, the followingmetricsandparametersaresetup:
1. The text ratioR in thewindow;
2. TheamountMof the fragments inuse;
3. TheheightHof thewindow;
4. ThewidthWof thewindow.
However, thesametechniquesareconsideredfor:
1. Themainbodyheightdetection[22], since itdoesnot require lineorwordsegmentation;
2. Theslantdetectionprocedure. Once theMfragmentshavebeenselected (Figure5), the slant
detectionalgorithm[9],describedinSection2, isappliedandtheslantanglesaredetected,one
per fragment. Themaximumandminimumslantanglesare ignoredaspossibleoutliers,while
slant isdefinedas thedetected slantof thepage. Theentiredocumentpage is thencorrected
accordingto theslantanglebyshiftingeachpixel so that
xf = x0+round [
y0 tan (
π slant
180 )]
(4)
yf = y0 (5)
where (x0,y0)defines the initialpositionof thepixeland(xf,yf) is thefinalpixelposition.
34
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