features of a binary image, ocr

So I have segmented an image consisting of letters, it's a binary image. I have used regionprops and bwconncomp, standard Matlab functions. So now I want to extract features on the region of pixels that are equal to one, into a feature vector. It is also good if the features does not depend on which position the region has, so if the regions is shifted in x or y direction then the same features will be obtained. Is there any Matlab function that can do that? Thank!

Answer 1

[Image Analyst]

Yes. regionprops().

[Adam]

Sir, can you give me some examples on how to use regionprops()? I have looked at it but I can't figure out what I should do with it, do you have any concrete ideas?

[Image Analyst]

A well commented tutorial is in my File Exchange: http://www.mathworks.com/matlabcentral/fileexchange/?term=authorid%3A31862 Look for Image Segmentation Tutorial.

[Adam]

Thanks! But do you have features that you can recommend in the case of getting a good hit-rate, i.e. the percentage of characters that are correctly found and classified? I have a benchmark program that can calculate the hit-rate which is nice if you want to know that your ocr-system is good or not. I don't really see how blobs can be convenient in an ocr-system, or maybe I'm wrong. I haven't tried it out yet.

[Image Analyst]

It's just a general purpose generic routine. It's not specially adapted for OCR. See http://www.mathworks.com/help/vision/ref/ocr.html#bt548t1-2_1

Answer 2

[Dima Lisin]

There is a built-in ocr function in the Computer Vision System Toolbox.