Cumulative results?

[karolzak]

Cumulative results?

[waiyeung1]

Hi @karolzak,
Thanks for the quick response and suggestion. Unfortunately, I can't share the documents.

I was able to determine the cause. With the configuration, the missing checkboxes actually had 2 rectangles drawn around them that weren't merged when group_countours() was called. As a result the checkbox wasn't considered a checkbox. By passing in epsilon=0.2, the 2 contours were merged and the checkbox was determined to be a checkbox.

Cheers,
Wai

[karolzak]

Hello @waiyeung1
Is it possible for you to share these documents?

Overall if the doc B is almost identical but smaller and more grainy it may require a different configuration than doc A.
E.g.: if for doc A, width range is (30,70) for doc B it should be smaller, like (20,50) (depending how much smaller is doc B. So for doc B I would try something like below:

# important to adjust these values to match the size of boxes on your image
cfg.width_range = [(20, 60)]
cfg.height_range = [(20, 60)]

# w/h ratio range for boxes/rectangles filtering
cfg.wh_ratio_range = [(0.8, 1.2)]

# num of iterations when running dilation tranformation (to engance the image)
cfg.dilation_iterations = [1]
cfg.dilation_kernel = [(1,1)]

But without seeing the actual documents - it is just guessing