Lightness inversion preserving colors

[mara004]

Hello,

I'm wondering how to invert lightness of an image while preserving colors, using Pillow?
e.g. white to black (and vice versa), dark green to light green, etc.
I am not looking for classical color inversion (which would, for example, turn dark green into light pink).

A demo of what I mean is Okular's "Invert Lightness" accessibility option.
For some context, I would like to implement something similar (a dark mode for PDFs), but with an option to exclude images from the inversion, which is relevant for machine-generated (not scanned) PDFs.

Thanks!

[Yay295]

I think you could convert the image to HSV and then invert the V channel.

[mara004]

Thanks for the suggestion, but this is not quite what I meant yet.
Inverting the V channel seems to keep colors as-is, but I still want to invert their lightness (e.g. dark blue to light blue).
Sorry about my wording, "preserving colors" is probably a bit misleading here.

[mara004]

Default	Invert Colors	Invert Lightness	Invert V Channel

[mara004]

I did some investigation: Okular's invertLightness() function (introduced with KDE/okular@aa68334) apparently uses HSL rather than HSV. However, I think HSL is not supported by Pillow yet?

[radarhere]

HSL isn't supported as a mode, no - https://pillow.readthedocs.io/en/stable/handbook/concepts.html#modes

You could use colorsys.rgb_to_hls and colorsys.hls_to_rgb to convert the values, then invert the lightness, then convert them back.

[homm]

Here is the full example using HLS color space:

import colorsys
from PIL import Image, ImageFilter


def invert_colors(r, g, b):
    h, l, s = colorsys.rgb_to_hls(r, g, b)
    l = 1.0 - l
    return colorsys.hls_to_rgb(h, l, s)


invert_filter = ImageFilter.Color3DLUT.generate(17, invert_colors)
Image.open('257923775.png').filter(invert_filter).save('257923775.out.png')

[mara004]

Thanks @homm and @radarhere for the advice!

I had found how to do it with OpenCV yesterday (and there's also a colour-science package that might work), but PIL and a built-in module of course fits better with my current dependencies.

[mara004]

@homm I'm afraid I don't yet understand the size parameter of ImageFilter.Color3DLUT.generate (the docs didn't help me much). How do you get to the value 17? Could you please explain?

[homm]

It’s a size of generated 3D lookup table. The bigger the size, the close interpolated values to the real ones. 17 is a safe minimal, less could be noticeable on some transformations. In your case I don’t think there would be any differences between 17 and higher sizes.

[mara004]

Thanks!