Submit bug report to skimage for offset case

[dchaley]

While working on #128 I found some discrepancies between my implementation and scikit's that I couldn't explain, when using a non-null offset. (aka recentering the footprint)

After a lot of head-banging, I actually think I found a bug in the scikit implementation.

Work:

• create a document showing the bug & why it's not the expected result
• submit to skimage: https://github.com/scikit-image/scikit-image/issues

[dchaley]

The bug report is complete. cc @lynnlangit

I'll take another look, it's probably good enough, then I'll submit an issue.

Ideally, I'd have an understanding how to fix the issue.
Then, I can compare my impl vs the fixed scikit impl.
But I'm not 100% sure how to fix it yet.

Not sure the logistics of merging an algorithm replacement that doesn't match??
Maybe I should be testing vs the opencv implementation, which seems to work. (It's also more obviously correct in principle, it's a much simpler algorithm.)

[lynnlangit]

nicely documented

use the simplest thing that works

[dchaley]

Bug report submitted: scikit-image/scikit-image#7315