Right now I have a label with many lines of text as below. I want to find a way to detect if there are any scratch in a line (the first and the last line). Any recommendation, thanks.

CodePudding user response：

Assumption the text lines are aligned horizontally.

• Threshold the black text to get a mask.
• Count the black pixel in each row of the image. This gives you a vector with the number of rows.
• In this vector, You should see a pattern of small and high values which represent empty and filled lines.
• Scratches are small values between two high values.

Another way is to fit a line using Hough transform: https://learnopencv.com/hough-transform-with-opencv-c-python/

CodePudding user response：

This is a pretty difficult issue since it is actually lack of print in a line than a strike-through.

Can you tell me the process you are following for OCR so that I can work in a solution?

My best guess right now is perform a Sobel operation in the horizontal direction and with text aligned. and if you find a line where there is no gradient change, then that is most likely a blank line and if you observe that such a blank line is very narrow and this line is intersecting/overlapping detected text, it is scratch that you are looking for.

But for me to give a better answer, I will need to know the steps you are following.