Incorrect warning during the check of the margin with invisible characters

[crux82]

In some papers, the SW incorrectly raises an error when checking the margin but there is no visible text there.

Actually, there is some text but it is invisible...

I think the problem is somewhere in formatchecker.py

                # Parse texts
                for j, word in enumerate(p.extract_words()):
                    violation = None
                    if float(word["top"]) < (57-self.top_offset):
                        violation = Margin.TOP
                    elif float(word["x0"]) < (71-self.left_offset):
                        violation = Margin.LEFT
                    elif Page.WIDTH.value-float(word["x1"]) < (71-self.right_offset):
                        violation = Margin.RIGHT

[crux82]

@ryancotterell Maybe I found a solution. Actually, it may sound like a workaround, but it seems quite impossible to robustly detect invisible characters: when a candidate area out of the margin is proposed, this is cropped and if all pixels are equal to the background, this is skipped.

You find this implemented in the branch