How to visualize an "ab" slice of CIE Lab with corresponding "L"?

[hminle]

Hi,

Do you guys have any idea on how to visualize an "ab" plane of CIE Lab with corresponding "L"?
Like what've been shown in this webpage: http://web.mit.edu/kenta/www/one/color-lab/dwvcvceg/color.html

Best,

[KelSolaar]

Hi @hminle,

We do not have anything readily available for that and while it is not easy to do that for arbitrary gamuts as it requires mesh geometry intersection, it is trivial to do for RGB colourspaces:

from colour.colorimetry import CCS_ILLUMINANTS, SDS_ILLUMINANTS
from colour.geometry import primitive_cube
from colour.models import Lab_to_XYZ, XYZ_to_Lab, XYZ_to_sRGB, sRGB_to_XYZ
from colour.plotting import XYZ_to_plotting_colourspace, artist
from colour.utilities import as_int, full, orient, tstack
from colour.volume import is_within_visible_spectrum, solid_RoschMacAdam

illuminant = 'D65'
sd_n = SDS_ILLUMINANTS[illuminant]
xy_n = CCS_ILLUMINANTS['CIE 1931 2 Degree Standard Observer'][
    illuminant]

vertices, faces, outline = primitive_cube(1, 1, 1, 32, 32, 32)
XYZ_vertices = sRGB_to_XYZ(
    vertices['position'] + 0.5,
    apply_cctf_decoding=False)
Lab_vertices = XYZ_to_Lab(XYZ_vertices, xy_n)

extent = np.max(
    np.abs([np.min(Lab_vertices[..., 1:]),
            np.max(Lab_vertices[..., 1:])]))
extent = np.around(extent * 1.1 / 100, 1) * 100

samples = 2048
L = 50

ii, jj = np.meshgrid(
    np.linspace(-extent, extent, samples), np.linspace(extent, -extent,
                                                       samples))
ij = tstack([ii, jj])

Lab_slice = full([samples, samples, 3], L)
Lab_slice[..., 1:] = ij

normal = [1, 0, 0]

XYZ_slice = Lab_to_XYZ(Lab_slice, xy_n)
RGB_slice = XYZ_to_plotting_colourspace(XYZ_slice, xy_n)
XYZ_mask = XYZ_to_sRGB(XYZ_slice, apply_cctf_encoding=False)
RGB_slice[np.any(XYZ_mask < 0, axis=-1)] = 1
RGB_slice[np.any(XYZ_mask > 1, axis=-1)] = 1

figure, axes = artist()
image = axes.imshow(
    RGB_slice,
    interpolation='bilinear',
    extent=(-extent, extent, -extent, extent),
    clip_path=None);

[KelSolaar]

I spent a bit of time refining the code for that, I might have something by the end of week that could be included in Colour. Stay tuned!

[hminle]

Thank you @KelSolaar :)

[hminle]

Hi @KelSolaar ,

In this case, how you can visualize the slice of another gamut, like "ProPhoto RGB"?
I tried to visualize both Lab slice of "sRGB" and "ProPhoto RGB" on the same figure.
But got the weird result.

The ProPhoto region is blue
The sRGB region is yellow.

But I think I made a mistake somewhere, since the ProPhoto region is smaller than sRGB.

[Myndex]

Hi @hminle

But I think I made a mistake somewhere, since the ProPhoto region is smaller than sRGB.

So I am going to guess that the vertical axis is ±b*, and the excursion you see in blue is common when you are mixing illuminants D50 and D65 without the appropriate transform.

sRGB is D65, and ProPhoto is D50. They don't mix (not directly) you need to adapt one or the other, and your LAB space needs to be adapted to that same whitepoint as well.

This "looks like" (taking a guess) that your LAB space is at D50, as well as ProPhoto, and that you brought in the sRGB without first doing a Bradford (or other) chromatic adaptation transform. If you want your LAB space to be in ProPhoto's D50, then you need to transform sRGB to D50. D50 is the illuminant used for the PCS in v2 and v4 ICC profiles.

Alternately, you could adapt your ProPhoto to D65, and use LAB with D65, and sRGB then would remain as D65.

[hminle]

Hi @Myndex ,

I convert everything to D50 illuminant.
Belows are my converting function.
The vertical axis is "a" channel, the horizontal is "b"

ILLUMINANT_D50 = colour.CCS_ILLUMINANTS['CIE 1931 2 Degree Standard Observer']['D50']

def srgb_to_Lab_D50(srgb):
    srgb_colourspace = colour.models.RGB_COLOURSPACE_sRGB.chromatically_adapt(ILLUMINANT_D50)

    XYZ = colour.models.RGB_to_XYZ(img_srgb.reshape(-1,3), 
                                         srgb_colourspace.whitepoint, 
                                         srgb_colourspace.whitepoint,
                                         srgb_colourspace.matrix_RGB_to_XYZ)
    Lab_srgb = colour.XYZ_to_Lab(XYZ, 
                                 illuminant=ILLUMINANT_D50)

    return Lab_srgb

def prop_to_Lab_D50(prop):
    prophotorgb_colourspace = colour.models.RGB_COLOURSPACE_PROPHOTO_RGB

    XYZ = colour.models.RGB_to_XYZ(img_prop.reshape(-1,3), 
                                 prophotorgb_colourspace.whitepoint, 
                                 prophotorgb_colourspace.whitepoint,
                                 prophotorgb_colourspace.matrix_RGB_to_XYZ)
    Lab_prop = colour.XYZ_to_Lab(XYZ, 
                                 illuminant=ILLUMINANT_D50)

    return Lab_prop

def lab_to_prop_D50(Lab):
    prophotorgb_colourspace = colour.models.RGB_COLOURSPACE_PROPHOTO_RGB
    XYZ = colour.Lab_to_XYZ(Lab, 
                            illuminant=ILLUMINANT_D50)
    
    img = colour.models.XYZ_to_RGB(XYZ, 
                 prophotorgb_colourspace.whitepoint, 
                 prophotorgb_colourspace.whitepoint,
                 prophotorgb_colourspace.matrix_XYZ_to_RGB)
    return img

def lab_to_srgb_D50(Lab):
    srgb_colourspace = colour.models.RGB_COLOURSPACE_sRGB.chromatically_adapt(ILLUMINANT_D50)
    XYZ = colour.Lab_to_XYZ(Lab, 
                            illuminant=ILLUMINANT_D50)
    
    img = colour.models.XYZ_to_RGB(XYZ, 
                 srgb_colourspace.whitepoint, 
                 srgb_colourspace.whitepoint,
                 srgb_colourspace.matrix_XYZ_to_RGB)

    return img

[Myndex]

Hi @hminle

Well if the vertical is ±a* then sRGB is exceeding ProPhoto in green, that's crazy.

Unfortunately I don't know Ken's code well enough to suggest a fix. But I'd suggest looking someplace where ProPhoto is being clipped.

[hminle]

Hi @KelSolaar ,

Is there anyway I can visualize 2 colour spaces (sRGB and ProPhoto RGB) in one Lab slice?
Maybe don't need RGB_slice, but only scatter points in ab space

[KelSolaar]

@hminle : The code has been merged in the develop branch!

[hminle]

Greatt, thank @KelSolaar . I will try it out.

[hminle]

So, the normalisation code you linked serves a slightly different purpose to that of the domain-range scale, the idea is that some of the colour models, e.g. ICtCp, have absolute scales so it becomes really hard to set 50% lightness with them, it might actually be something like 0.027 or alike. If normalise is on, and you use 0.5 for the origin, the section will be in the middle of the data along the chosen axis.

When you say image pixels, would you need to add a scatter point of image data on-top of the section(s)?

Hi @KelSolaar , yes. I want to add scatter point of image data on--top of the sections. Just want to see out-of-gamut pixels

[KelSolaar]

Try something like that maybe: https://colab.research.google.com/drive/1bljhLAfzm-36GQfjALYrQdHuf7fNYTaw#scrollTo=oAwadLyX1GQB

[hminle]

Thank @KelSolaar , I've just requested an access to your link.