Open source engines from Vivliostyle, Weasyprint, AthenaPDF
More implementations for print than for screen
PDF has been doing color properly since 2000 (PDF 1.3)
Why should we listen to you?
CTO of bfo.com; working with PDF since 1999
Member of CSS Working Group, various PDF Association working groups
Knows just enough about color to be dangerous
We have implemented CSS Color 4 for PDF output
Let's compare: CSS Color 3
CSS
PDF
sRGB
DeviceRGB
HSL
-
ICCBased (also CalRGB, CalGray)
Let's compare: CSS Color 4
CSS
PDF
sRGB
DeviceRGB
HSL
-
HWB
-
ICC (also display-p3, rec2020, xyz etc)
ICCBased (also CalRGB, CalGray)
device-cmyk
DeviceCMYK
Lab
Lab
LCH
-
-
DeviceN (also Separation)
Alternative coordinates: HWB, HSL and LCH
LCH and Lab are different views of the same colorspace.
Lab uses cartesian coordinates, LCH uses polar. HWB/HSL
are roughly the same, but for sRGB.
lab(75% 40 30) = lch(75% 50 36deg)
Gradients
Simply converting the endpoints from LCH to Lab is not enough.
We also need to control how we interpolate.
lch(75% 70 -135deg)lch(75% 70 45deg)
lab(75% -50 -50)lab(75% 50 50)
Note: macOS preview may not display these gradients properly
Interpolation
In CSS, all interpolation is linear. But PDF has more choices:
we can stitch functions, use sampled functions, even PostScript.
So simulating an LCH gradient in Lab is easy; we just need the right function.
Many ways to do this: we make a linear function, sample it halfway
and measure ΔE(CIE94). If it's too far, we split the function and
repeat for each half - De Casteljau's algorithm. Good for simulating
HSL/HWB in RGB too.
Let's recompare: CSS Color 4
CSS
PDF
sRGB
DeviceRGB
HSL
DeviceRGB
HWB
DeviceRGB
ICC (also display-p3, rec2020, xyz etc)
ICCBased (also CalRGB, CalGray)
device-cmyk
DeviceCMYK
Lab
Lab
LCH
Lab
-
DeviceN (also Separation)
So it's all working perfectly?
ICC profiles of input-type Lab and XYZ are disallowed in PDF.
So we can't use an
identity XYZ ICC profile
for color(xyz n n n). Same for 7-color
FOGRA55.
Named Color ICC profiles also disallowed. Exceedingly rare.
DeviceN/Separation colors are the better solution; Maybe css-color-n (for n≥5)?
2D Coons Patch gradients are required for CSS
conic-gradient.
But these typically interpolate in the wrong color space - sRGB or similar.
The corner colors are correct; the solution is to use smaller patches so the
interpolation matters less.
Coons Patches will also be required if the proposed
<meshgradient>
is added to SVG.
Note: macOS preview may not display this gradient properly
CMYK is widely supported. All commercial engines use cmyk(0%, 22%, 5%, 0%)
ICC may be supported, in implementation-specific ways.
@color-profile will help.
Device-independent color is required for PDF/A and PDF/UA, widely required by governments.
Demand for these should help adoption.
