Iteratively rotate a line in the text, then project is on the horizontal axis. The rotation giving you the highest values in the projection, will be the (minus) slant.
Correction is then of course: rotate the text with minus slant.
I don't know Joachim, but I think you are detecting the deviation of lines not running horizontal. Slant is (when your correction has been made, i.e. when the text lines run horizontal) how much the characters lean forward or backward; a property of the font.
Your correction should be performed first in order to be able to select one textline.
However, I would not use the Hough transform (computationally expensive and no clear peaks in the accumulator because text-lines are thick). I would iteratively rotate the whole text and project it on the vertical axis, and proceed as I described for detecting slant of characters.
@Joachim Pimiskern , dear sir, i am talking about the slant in the writing not the skew in the whole image. slant is the angle with which the characters are written in a line and is with respect to the vertical line. while skew is the angle of the whole image with respect to a base line.
@Javed : did YOU downvote Joachim's contribution? ~!@#$%^
People are only trying to help, and have put an effort. Please, only use that feature for racist remarks and such.
@Joachim : floodfilling closed areas within characters and determining that ratio would yield how stretched characters are, not their tilt.
Anyway, floodfill needs to be seeded. That would be a semi-automatic method unless more complicated image-processing is done to generate the seeds.
- Badges
- Science topic
- Similar topics
- Engineering
- Electrical Engineering
- Signal Processing
- Image Processing