[compositing] Question: in what color space will blending occur? #427
Comments
|
As you say, for the legacy sRGB syntaxes we are limited by Web compat to do the wrong thing (operating directly on gamma-encoded values). Blending/mixing/interpolation can have an opt-in to do better things (either linear-light for physical light modelling, or perceptual uniformity, or chroma-preserving perceptual uniformity). For the non-legacy Wide Color Gamut stuff, converting to sRGB first is absolutely not the way to go as the Interpolation section makes clear, in particular Color space for interpolation. The work to add these opt-ins for transitions, animations, gradients, and compositing is ongoing. |
svgeesus
changed the title
|
According to #517 (comment) there already exist some browsers on macOS which perform compositing in the So the "status quo" statement would be to say that "legacy" compositing/blending happens in an implementation-defined colourspace, which will typically be the colourspace of the system's display output. The specification should also note that "legacy" compositing/blending is performed without gamma correction (i.e. without first converting to a linear light or a perceptually uniform colourspace). Note that for "Simple alpha compositing" (where the alpha channel of a sample is a value indicating the ratio within the sample coverage area of opaque colour vs full transparency), performing the compositing in linear light is generally considered to be correct. The blending of opaque colours in other blend modes is typically used for artistic effect, and the colour space that they are performed in should be selectable since it will give noticeably different results; a non-linear or perceptual space might be preferred over linear. |
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN" "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<title>CSS Test: hsla()</title>
<link rel="author" title="L. David Baron" href="https://dbaron.org/" />
<link rel="author" title="Chris Lilley" href="https://svgees.us" />
<link rel="author" title="Mozilla Corporation" href="http://mozilla.com/" />
<link rel="help" href="http://www.w3.org/TR/css3-color/#hsla-color" />
<link rel="match" href="t425-hsla-basic-a-ref.xht" />
<meta name="assert" content="Test basic functioning of hsla() colors." />
<style type="text/css"><![CDATA[
html, body { background: white; }
div { width: 12em; height: 6em; }
#one { background-color: hsla(120, 100%, 70%, 1.0); margin-bottom: 0 }
#two { background-color: hsla(120, 100%, 50%, 0.6); margin-top:0 }
]]></style>
/*
hsla(120, 100%, 70%, 1.0) is rgb(40% 100% 40%)
hsla(120, 100%, 50%, 0.6) is rgb(0% 100% 0% / 0.6)
composite rgb values at 60% over white in sRGB = rgb(40% 100% 40%)
BUT if compositing is done with used values, in device RGB
you will get a mismatch on WCG screen
because the red and blue components will not be zero
and the green will not be 100%
*/
</head>
<body>
<p>Test passes if you see a light green square.</p>
<div id="one"></div>
<div id="two"></div>
</body>
</html>reference is <!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.1//EN" "http://www.w3.org/TR/xhtml11/DTD/xhtml11.dtd">
<html xmlns="http://www.w3.org/1999/xhtml">
<head>
<title>CSS Test: hsla()</title>
<link rel="author" title="L. David Baron" href="https://dbaron.org/" />
<link rel="author" title="Chris Lilley" href="https://svgees.us" />
<style type="text/css"><![CDATA[
html, body { background: white; }
div { width: 12em; height: 12em; }
#one { background-color: hsla(120, 100%, 70%, 1.0); margin-bottom: 0 }
]]></style>
</head>
<body>
<p>Test passes if you see a light green square.</p>
<div id="one"></div>
</body>
</html> |
Right now, blending occurs in sRGB space since everything is in sRGB space. Admittedly this is bad from a physical modeling point of view for light (a nice text to read is https://developer.nvidia.com/gpugems/gpugems3/part-iv-image-effects/chapter-24-importance-being-linear) for perceptual as-is it makes some sense.
But what happens with stuff in a different color space? Should the color be converted to sRGB first then blended or something else?
The text was updated successfully, but these errors were encountered: