Riadh Fezzani

Modifying `mode` is, I think, the best approach to avoid searching for image minimum value.

> My suggestion is to divide this issue into 2 PRs, one for the 11 flatten matches, that can be implemented and merged to main first, and a second one...

Sorry for coming late to the discussion. I agree with your last comment @jmetz, I refactored the canny function in #4342 ;-). I see two other options concerning the discussion...

@jmetz I propose in #5114 to deprecate `low_threshold` and `high_threshold` in favor of the unique parameter `thresholds`, this may ease the implementation of this PR :wink:

I confirm, I can reproduce: ## dev branch ``` 3.8.2 (default, Mar 26 2020, 15:53:00) [GCC 7.3.0] Linux-5.13.0-7614-generic-x86_64-with-glibc2.10 scikit-image version: 0.20.0.dev0+git20220725.38b595d60 numpy version: 1.19.2 Shift (binary image) : [-5. -3.]...

Well after some investigation, setting `normalization` parameter to "None" fixes the issue. Reading the note in the [doc](https://scikit-image.org/docs/stable/api/skimage.registration.html#phase-cross-correlation) > In a high noise scenario, the unnormalized method may be preferable.

I agree that the normalization by default broke the backward compatibility :grimacing:

Please see #5459 and #5460 for cross ref

>Tried to fix docstring couple of times already, not sure where the unindent is happening Have you tried the solution proposed [here](https://stackoverflow.com/questions/38761546/markup-ending-without-empty-line-but-markup-is-only-thing-in-comment)?

Thank you @katerina-ds for raising this, and sorry for the answer delay. @lagru > Nevertheless this seems like behavior that should either be addressed in the documentation or more likely...