Panel has 10 sides instead of 4

[robstolarz]

While running ./ohnoify.py ./build/originals/media/tumblr_owj0qjOPZ01vbwf2ko1_1280.png -o /tmp --debug with the comic from http://webcomicname.com/post/165508600064, this debug output is produced:

[(10, 152484), (4, 150926), (4, 149362), (9, 736), (9, 550), (9, 486), (8, 468), (9, 432), (9, 420), (7, 390), (12, 368), (7, 357), (12, 352), (5, 348), (10, 342), (11, 336), (10, 336), (8, 320), (11, 315), (13, 300), (6, 297), (12, 286), (5, 285), (8, 272), (9, 264), (11, 252), (10, 247), (10, 247), (12, 240), (8, 238), (12, 234), (8, 210), (8, 210), (9, 208), (10, 208), (8, 195), (12, 180), (8, 154), (8, 144), (10, 135), (9, 135), (8, 117), (4, 36), (8, 30), (8, 25), (7, 20), (7, 20), (6, 20), (2, 13)]
image had 2 rectangles.
of these, the closest was 88.83692925805124 units from the bottom right

I'm on Arch Linux 5.0.7-vfio, with Python 3.7.3 and 2.7.16, OpenCV 4.1.0.

[robstolarz]

From the interactive debug, it seems like the first panel is detected as having ten sides because it isn't detected as fully connected at the top left corner. It doesn't prevent the panel we want from being detected and clipped, but when ohnoify.py is run normally this causes the panel count to be under 3 so the program will decide it's just not a comic.

Theories for fixing this:

1. currently the contour detection is augmented by some cheap threshold ops, maybe replacing this with Canny with a high aperture is a better bet