[Python] How to remove dotted boundary boxes while preserving handwritten digits/characters...

I'm working with images that contain handwritten digits and letter 'x' (signifying that that box should be considered 'empty') where each character is written inside a dotted box. I'm trying to remove those dotted boxes around the characters for later ocr. Additional issue is that the images are of varying quality and quite frequently those dotted lines merge into solid ones. I've tried isolating only characters which more often than not overlap with those boxes but got nowhere.

Here are examples of images im working with:

Example 1

Example 2

Example 3

Example 4

Somewhat successful pipeline so far is (with python and openCV):

1. greyscale
2. otsu
3. erode & dilate
4. ideally there should be left only 4 characters

However the handwriting in some cases is pretty faint and during erosion characters get eroded as well. I've also tried dilating images and drawing boxes by detecting horizontal and vertical lines, and removing them after, but since lot of characters either overlap with the box or go beyond it, this approach too has yielded pretty questionable results.

I'm not sure how to proceed or how to approach this, so any suggestion is more than welcome, thanks in advance.

Continue reading...