[Python] How to remove overlapping noise lines without damaging characters when using Hough...

I’m working on extracting text from synthetic images like this:



When a noise line overlaps a character (for example, a line crosses the letter P or the @ symbol), removing the line also removes parts of the character underneath. This causes OCR to fail or misread the text.

I have tried:

Using cv2.HoughLinesP to detect the lines.

Drawing over the detected lines or applying inpainting to remove them.

Continue reading...