Help on character segmentation for license plate recognition

Hi all,

currently I am working on a code for license plate recognition, I have finished coding for license plate extraction,I need help for character segmentation into individual character for some of the connected character inside the car plate.I'm thinking to use FFT2 to improve the edge of the image and then inverse it back using IFFT..is that possible to do this??

output image:

http://s1202.photobucket.com/albums/bb375/ShuWei88/

Second image after thresholding

http://i1202.photobucket.com/albums/bb375/ShuWei88/w2-1.jpg

[Jonathan Ocampo]

Hello Catherine couldyou help me with licenseplate extraction code

Answer 1

[Sean de Wolski]

You know how wide a character is, I would begin with that knowledge to indiscriminately segment the characters from each other.

1. start at left edge, set all values in the column x_length over to black, repeat until edge of that connected component.
2. repeat with next CC etc.

[Catherine]

But i got different database with different width and height of the character..is that fine to use this method?

CC stand for character?? Anyway I'm sorry for asking this silly question..new user in matlab central..perhaps you dont mind

[Sean de Wolski]

CC = connected component.

Are there always seven characters? Are they always at a small angle? To make a completely generalized license plate recognition system is not a trivial task. For this image, a simple threshold and knowledge of number of characters and/or size of each character should be all you need. Follow that with a skeletonization (doc bwmorph) and the characters should be all set. For other images you may need more complicated methods.

Welcome to Matlab Answers!

[Catherine]

not always,my database have six or seven characters inside the image.The majority is 7 characters.Yes there are always at small angle.

http://i1202.photobucket.com/albums/bb375/ShuWei88/w2-1.jpg

Here is the image after some thresholding process when i cropped my detected car plate region and filtering of small particles..skeletonization??what function does it involve??this is the image after skeletonization...thanks =)

http://i1202.photobucket.com/albums/bb375/ShuWei88/untitled.jpg

[Catherine]

Hi sean de!i have solve the segmentation problem for the connected characters by using width and height ratio and its work.Did u know how to combine the different output image which are same size into a single image??i got 6 sub image and i wish to combine them into 1,hope u can help...

thanks

Answer 2

[Image Analyst]

No, I don't think that edge enhancement should be necessary. Why do you think it should? I think you need to study on on OCR algorithms first to see how the pros do it: http://iris.usc.edu/Vision-Notes/bibliography/contentschar.html#OCR,%20Document%20Analysis%20and%20Character%20Recognition%20Systems