[WIP] Restore the proper scale of Lab and Lch

[Ogeon]

This restores the scale of Lab and Lch to be the same as in all of the specifications within reach. That is:

• Lab::l and Lch::l goes from 0 to 100,
• Lab::a and Lab::b goes from -128 to 127,
• Lch::chroma goes from 0 to about 128-181.

This is a breaking change, since everything that uses these values directly will be wrong. Closes #49.

[Ogeon]

The tests are still wrong and needs to be tweaked. Some of the inaccuracies are huge, like

    left = Lab { l: 91.11652, a: -51.481186, b: -14.387321 }
    right = Lab { l: 91.12, a: -48.08, b: -14.14 }

[sidred]

I've used a tolerance of 0.05 which translates to 5 when scaled.

I wonder if this is due to the differences in the default white point being used

pub const X_N: f64 = 0.95047;
pub const Y_N: f64 = 1.0;
pub const Z_N: f64 = 1.08883;

[sidred]

Nope. Color mine uses the same values:

        final double REF_X = 95.047; // Observer= 2°, Illuminant= D65
        final double REF_Y = 100.000;
        final double REF_Z = 108.883;

But the formula in color mine is incorrect

        final double REF_X = 95.047; // Observer= 2°, Illuminant= D65
        final double REF_Y = 100.000;
        final double REF_Z = 108.883;

        double x = pivotXyz(xyz.X / REF_X);
        double y = pivotXyz(xyz.Y / REF_Y);
        double z = pivotXyz(xyz.Z / REF_Z);

        return new Lab(116 * y - 16, 500 * (x - y), 200 * (y - z));

Its not using the transformation function

        fn convert<T: Float>(c: T) -> T {
            let epsilon: T = (flt::<T,_>(6.0 / 29.0)).powi(3);
            let kappa: T = flt(841.0 / 108.0);
            let delta: T = flt(4.0 / 29.0);
            if c > epsilon {
                c.powf(T::one() / flt(3.0))
            } else {
                (kappa * c ) + delta
            }
        }

I guess we need a better set of tests

[sidred]

We have a test for Xyz <-> Yxy, already from the CIE tables.

https://www.rit.edu/cos/colorscience/rc_useful_data.php has a few calculated values for Xyz, Lab and Lch ( CIEDE200 Color Difference formula, Pointer data). But it uses a 10 degree observer.

This should cover the CIE spaces.

http://stackoverflow.com/questions/3620663/color-theory-how-to-convert-munsell-hvc-to-rgb-hsb-hsl

points to http://www.cis.rit.edu/research/mcsl2/online/real_sRGB.xls which has rgb to xyz and yxy conversions.

The rgb variants conversions (hsl, hsv, hwb etc.) need to be figured out

Just adding it here for reference if the tests need to be updated.

[Ogeon]

Thanks, I'll take a look at them later and see if it's possible to piece things together.

[homu]

☔ The latest upstream changes (presumably #56) made this pull request unmergeable. Please resolve the merge conflicts.

[Ogeon]

Turns out the accuracy of the Yxy values were too low, so I'm currently working my way through them, adding more decimals.

[sidred]

So the actual scaling part is done? Is it just the tests remaining?

[Ogeon]

Yes, unless I forgot something. I'm pretty much just copy-pasting XYZ numbers to the easyrgb calculator, dividing them by 100, converting, and copy-pasting the xyY numbers back.

[sidred]

Closed by #61