how to segment connected letters in a word by using its vertical profile?

Hi..

I would like to segment connected letters in a word by using the vertical profile of that word.

I've produced the Vertical profile of that word & I need to segment the letters to separated images.

please can any one help me??

there will be no thinning to the word image the only thing I apply is im2bw().

Answer 1

[Walter Roberson]

Ideally, you put boundaries where the profile goes from 0 to non-zero, and from non-zero to 0.

Answer 2

[Image Analyst]

You can sum the image horizontally, across columns, to get the vertical profile. Then you can threshold that to get the lines of text. I believe I did that for you or someone a few weeks ago. Once you have a line of text, you can look for word separators (spaces) by looking at the horizontal profile (sum vertically along rows). The spaces between words will be large and the spaces between characters in a single word will be smaller. That algorithm seems fairly straightforward to program up. Is that what you were wanting to do? If so, start programming.

[s]

Hi.. Thank you all about what you said..

actually.. my words contain Arabic connected letters and I need to separate those letters i.e. producing images each image containing a single letter.

here is what I did till know

-----------------------------------

img = imread('test.png'); img=im2bw(img);

sumAlongColumns = sum(~img, 2); %vertical profile%

-----------------------------

then, I could not continue :(

[Image Analyst]

This is where you continue: http://iris.usc.edu/Vision-Notes/bibliography/char1032.html#Arabic%20Character%20Recognition. There you will find many papers, perhaps hundreds of them, detailing successful algorithms that you can follow.

[s]

thank you so much about the link..

actually I already decided my algorithm but still could not segment my characters by using the vertical profile

[Image Analyst]

Then choose a different algorithm - one that works this time. That link has hundreds of them.

[s]

I am sorry if I were not clear..

the algorithm is true, but do not know how to use Matlab to accomplish this task :(

[Image Analyst]

http://www.mathworks.com/matlabcentral/answers/8026-best-way-s-to-master-matlab

[s]

thank you so much about trying to help :)