uint8 overflow in exposure.adjust_gamma

[florent-cd]

Description

When using the function skimage.exposure.adjust_gamma with an uint8 image with a gain superior to 1, the integers can overflow and return false results. A value bigger than 255 should be set to 255.

Way to reproduce

from skimage.exposure import adjust_gamma
import numpy as np

img = 255 * np.ones((3, 3), dtype=np.uint8)
transformed_img = adjust_gamma(img, gamma=1, gain=1.1)
print(transformed_img[0, 0])

This returns a value of 24 which is wrong.

The problem is in the definition of the lut in the function _adjust_gamma_u8 (l443):

    lut = (255 * gain * (np.linspace(0, 1, 256) ** gamma)).astype('uint8')

Before recasting as 'uint8' a max(255, lut) should probably be applied.

Version information

# Paste the output of the following python commands
from __future__ import print_function
import sys; print(sys.version)
import platform; print(platform.platform())
import skimage; print("scikit-image version: {}".format(skimage.__version__))
import numpy; print("numpy version: {}".format(numpy.__version__))

3.9.5 (default, May 18 2021, 19:34:48) 
[GCC 7.3.0]
Linux-4.15.0-143-generic-x86_64-with-glibc2.27
scikit-image version: 0.18.1
numpy version: 1.20.2

[hmaarrfk]

Thank you for the very detailed and complete bug report.

Would you be willing to provide the pull request to fix this issue?

I would suggest you include your failing example as a test and assert that the result is correct.

[florent-cd]

Yes, I can do the PR (probably this week-end)