I have searched about this field and I found some papers that present new methods to extracting texts from images, but I have a grayscale image consists of a simple background and some texts.so I need a method that everyone works with it. please provide details on how this can be done.
Asked
Active
Viewed 1,516 times
2
-
1Maybe have a look to [this](http://stackoverflow.com/questions/19960826/how-to-make-the-blackboard-text-appear-clearer-using-matlab/19962240?noredirect=1#comment29743737_19962240) – Vuwox Nov 14 '13 at 16:18
-
[The article](http://www.math.tau.ac.il/~turkel/imagepapers/text_detection.pdf) – Vuwox Nov 14 '13 at 16:19
-
for example "Text detection and recognition in natural images" by Steven Bell that presents a method without performing segmentation. – top.eng Nov 14 '13 at 16:21
-
this article seems to be for natural images only but my images as i said have very simple background for example a red background with probably some noises. do you think this paper will help me? – top.eng Nov 14 '13 at 16:29
-
@top.eng You should try reading the paper describing the stroke width transform. If it works for complicated backgrounds it should work in your case as well. – user2469775 Nov 14 '13 at 16:44
-
1If you look to the first link I send, the guys use this paper on a blackboard, as you, it a uniform background. Have a look closer, I don't know if it good for you or not. – Vuwox Nov 14 '13 at 16:44
-
1@top.eng morover since you already have an implementation of SWT - why don't you just give it a try? – user2469775 Nov 14 '13 at 16:44
-
thank guys.I will read the paper... – top.eng Nov 14 '13 at 16:59
1 Answers
2
Here an article about text segmentation.
the article
And here an easy way to segment your image in 2 class.
I = imread('...'); % Your board image
ThreshConstant = 1; % Try to vary this constant.
bw = im2bw(I , ThreshConstant * graythresh(I)); % Black-white image
SegmentedImg = I.*repmat(uint8(bw), [1 1 3]);
Just do imshow(bw);
and you will have a 2 color image normally well segmented.
If the threshold is too strong, try to turn around 0.5 to 1.5 with ThreshConstant
.

Vuwox
- 2,331
- 18
- 33
-
I tried your answer for [this question](http://stackoverflow.com/users/2049859/top-eng?tab=reputation) and helped me. if you post it here i want to mark it as my answer. – top.eng Nov 14 '13 at 19:53
-
-
You're welcome. But this is a simple ostu 2-class segmentation. If you maybe look for more class segmentation. This [file exchange](http://www.mathworks.com/matlabcentral/fileexchange/26532-image-segmentation-using-otsu-thresholding) is really good. – Vuwox Nov 14 '13 at 20:01