Detect presence of text

asked 2015-03-06 07:46:49 -0600

I'm trying to extract signs from images and classify them. Extracting the signs and doing some classifications already works quite good.

Now I struggle with the simple classification if the sign is completely empty or contains some text. I don't need to do OCR on the text or anything related I just need a simple measure to decide between those two classes. One issue is dirt on the signs I have to classify. Simple thresholding and counting black vs. white pixels doesn't work because pixels counts are sometimes very close.

What would be a good approach to start classification in my case?

edit retag flag offensive close merge delete

Comments

1

Not sure if i've understood clearly your task..but you have to check whether an area is blank or not? If the area is white, since you mentioned black&white count, why not check the average pixel value? Any example image you're dealing with?

David_86 gravatar imageDavid_86 ( 2015-03-06 08:07:15 -0600 )edit

Blank in my case could also mean dirty but no text. Sometimes a completely empty but dirty sign ends up with black regions after thresholding that produce more black areas than regular text.

andy8080 gravatar imageandy8080 ( 2015-03-06 08:13:03 -0600 )edit