Added support for binary images in is_low_contrast to fix imshow of binary images

[MadsDyrmann]

Description

imshow stopped working when plotting boolean images, since np.percentile no longer support boolean arrays. This PR makes sure that is_low_contrast does not fail when the input is a boolean array.

This is a small test-script that would fail before this change:

import numpy as np
from skimage.io import imshow

im = np.ones((10,10), dtype=np.bool)

imshow(im)

Checklist

• Docstrings for all functions
• Gallery example in ./doc/examples (new features only)
• Benchmark in ./benchmarks, if your changes aren't covered by an
  existing benchmark
• Unit tests
• Clean style in the spirit of PEP8

For reviewers

• Check that the PR title is short, concise, and will make sense 1 year
  later.
• Check that new functions are imported in corresponding __init__.py.
• Check that new features, API changes, and deprecations are mentioned in
  doc/release/release_dev.rst.

[pep8speaks]

Hello @MadsDyrmann! Thanks for updating this PR. We checked the lines you've touched for PEP 8 issues, and found:

• In the file skimage/exposure/exposure.py:

Line 651:39: E712 comparison to False should be 'if cond is False:' or 'if not cond:'

Comment last updated at 2021-02-23 16:44:59 UTC

[jni]

Thanks @MadsDyrmann!

Very happy to merge this fix. Do you want to add a test to make sure it doesn't break in the future?

Having said all this, in the medium term (~1y) we aim to pull out the matplotlib dependency and make sure the focus of scikit-image is on image processing, not display or IO. So you should probably start migrating your downstream code to start using plt.imshow directly.

[mkcor]

So you should probably start migrating your downstream code to start using plt.imshow directly.

May I recommend plotly.express.imshow as an alternative? With plt.imshow I ran into the issue of being unable to specify value ranges for multichannel data. See this gallery example for basic px.imshow usage.

[jni]

@mkcor well, plotly plots it in a web browser, which some people are allergic to—cough cough cough. 😜

[mkcor]

@jni just an alternative... 😜

[emmanuelle]

Maybe I'm missing something, but if a boolean image has both True and False values, then the function should return False, shouldn't it? I have the impression that the returned value should be

Suggested change

	return not(image.any() and (image == False).any())

[emmanuelle]

anyway a test (as @jni suggested) could either prove me that I'm wrong :-) or check that the behaviour is as expected

[rfezzani]

When image.dtype == bool, is_low_contrast is True when the image is constant: all filled with True or with False, ie when False and True are not both in image.

So I think that you are right @emmanuelle!

BTW, I think that this test must take place at the beginning of the function, even before checking for image dimensions, because rgb2gray will convert the image to floating point...

[emmanuelle]

@MadsDyrmann are the different suggestions clear for you? Thanks again for the PR!

[mkcor]

Hi @MadsDyrmann,

To add a test, edit file skimage/exposure/tests/test_exposure.py: Line 749, right after test function test_is_low_contrast(), you can define another test, say, test_is_low_contrast_bool_image() and use the example you came up with, im = np.ones((10,10), dtype=np.bool).

Hope that helps!

[alexdesiqueira]

Hi everyone,
@MadsDyrmann, would you like to continue working on this PR? We could help you finishing it, if you'd prefer.
Thank you very much for your help so far!

[grlee77]

@MadsDyrmann, I have resumed this PR in #5467 so am closing this one as outdated.

Please confirm your name and email address either here or in that PR if you want to be attributed as a co-author on the commits.