Fix f_lab accuracy

[iynehz]

This fixes GH-6.

[mohawk2]

Thank you for this! I've enabled the CI, as you might see.

Putting xyz(100%, 100%, 100%) into https://products.aspose.app/svg/color-converter/xyz-to-lab gives lab(98.291%, 1.146%, 2.521%), while your updated test shows pdl(1,1,1)->apply($t) gives pdl(100, 8.5945916, 5.5564131). Can you help me understand the discrepancy?

By the way, I'm thinking to change this module from its own Perl-level calculations to using PDL::Graphics::ColorSpace, which has both C/PP implementations of the various functions, and more tests with more values including the ones you mentioned on #6. Do you have an opinion on that?

[mohawk2]

At the very least we'll want to copy over all the test values so that we have a measure of assurance of correctness.

[iynehz]

while your updated test shows pdl(1,1,1)->apply($t) gives pdl(100, 8.5945916, 5.5564131). Can you help me understand the discrepancy?

Not sure but at least we align with Matlab. If you try it on their web: https://www.mathworks.com/help/images/ref/xyz2lab.html

[mohawk2]

while your updated test shows pdl(1,1,1)->apply($t) gives pdl(100, 8.5945916, 5.5564131). Can you help me understand the discrepancy?

Not sure but at least we align with Matlab. If you try it on their web: https://www.mathworks.com/help/images/ref/xyz2lab.html

The image shows (100, 8.5385, 5.5939). It's not a big discrepancy, but that's different from the converter utility you linked me to on #6, and both the old and new values in your updated test. The word "align" feels like it's adding confusion, not taking it away?

[mohawk2]

And to add to the fun, P:G:ColorSpace shows:

pdl> use PDL::Graphics::ColorSpace
pdl> p xyz_to_lab(pdl(1,1,1), 'sRGB')
[100  8.5422401  5.5990201]

[iynehz]

The image shows (100, 8.5385, 5.5939). It's not a big discrepancy, but that's different from the converter utility you linked me to on #6, and both the old and new values in your updated test. The word "align" feels like it's adding confusion, not taking it away?

Well I am no expert in this area. What I can do is largely about comparing the library's results vs what's from other calculators. In this case MATLAB is obviously more trustworthy than the other one.

[mohawk2]

As a bit of fun I thought I would try the MATLAB stuff in Octave, which gave the same results (after installing the relevant package):

$ octave
[snip]
octave:1> pkg install -forge image # if needed
[snip]
octave:2> pkg load image
octave:3> xyz2lab([1 1 1])
ans =
   100.0000     8.5385     5.5939