7 ms·

Color Spaces

236 points by tylerchr 8y ago

platz 8y ago
gimp needs a LAB mode
- pmoriarty 8y ago
  It has one.
  - platz 8y ago
    i see. it's just operates very differently; i was hoping to view the result of applying LAB curves in real time on the original image, instead of a multi-step process of decomposing and re-composing each time to view the results. I guess I can just use two windows though
teddyh 8y ago
See also the venerable Color FAQ, a link to which is conspicuously absent: http://www.poynton.com/ColorFAQ.html http://www.poynton.com/ColorFAQ.html
pzone 8y ago
The CIE XYZ color space is fundamental to all modern color space processing. Very cool that it was defined in 1931 and has held up more-or-less unchanged to today. Be sure to click and drag on the 3d cubes!
- zokier 8y ago
  > Very cool that it was defined in 1931 and has held up more-or-less unchanged to today I wonder if the color mixing experiments have been recreated since the 30s, and if there has been any observable shifts in the curves
  - jacobolus 8y ago
    There have been a bunch of additional experiments. But under well-lit conditions with typical light sources, for a typical viewer with normal color vision looking directly at an image / colored object, the 1931 functions are close enough that switching isn’t worth the transition cost. The best recent data can be found at http://www.cvrl.org/ http://www.cvrl.org/
mroche 8y ago
This is a very well written article, and boils down the complexity of color science into very digestible bites. As someone who has to deal with color spaces daily (VFX/CG), having something like this when I started would have been incredibly helpful.
Straw 8y ago
No mention of CIELAB? Unfortunate, its a really neat way to get perceptual uniformity, has intuitive coordinates.
kuon 8y ago
For web design, I use HSLuv[1] which is very handy. [1]: http://www.hsluv.org/ http://www.hsluv.org/
- misterdoubt 8y ago
  That's damn cool.
- Exuma 8y ago
  Can you explain in laymans terms to another web designer why that is important? Like what problem does it solve?
  - kuon 8y ago
    For example, you want to dynamically generate a background and text color based on a hue. With hsluv you can simple do `hsluv(120, 60%, 10%)` for background color and `hsluv(120, 60%, 90%)` for foreground text color, you will have text on background with 80% perceived luminance difference, you can replace the `120` in my example with whatever hue in the `0-360` range and have consistent results. The idea is that the color space is linear (perceived) with saturation and luminance regardless of hue. You can see it clearly if you compare something like http://www.hslpicker.com http://www.hslpicker.com with http://www.hsluv.org/ http://www.hsluv.org/ if you drag the hue slider on the hsl version, the luminosity varies, it does not on the hsluv version.
    - Exuma 8y ago
      Ahh, very interesting. That is cool.
jacobolus 8y ago
Like many resources about color, this one has an output-medium-centric followed by early-20th-century-experiment-design-centric viewpoint, presumably because that was where the author started and what he read about. But however perfect the exposition might otherwise be, that is the wrong way to start a general discussion of the subject, and the result is inevitably misleading and confusing for non-experts. > it is a pragmatic approach Having tried to talk face-to-face to many people from different backgrounds about color, in practice this doesn’t work. People are starting with too many misconceptions and gaps in foundational knowledge. If you want to understand color you have to start with at least the basics about the physiology of human vision, along with some basic optics. It is all but impossible to properly understand what is going on with RGB displays, understand what “white point” means, etc. without that background. One of the resource recommended at the end, Bruce MacEvoy’s site http://handprint.com/LS/CVS/color.html http://handprint.com/LS/CVS/color.html is a much better starting place, albeit pretty long for someone just interested in the basics. Or there are many books about human color vision and color reproduction, some friendlier for laypeople and others highly detailed and technical. To understand recent color models Mark Fairchild’s book Color Appearance Models is pretty good, and the first half is a pretty accessible general overview. Or for more about color reproduction try Billmeyer and Saltzman’s Principles of Color Technology (recent editions rewritten by Roy Berns). For more about colorimetry try Hunt & Pointer’s book Measuring Color. Or for something historical, try Kuehni’s Color Space and Its Divisions.
- cmurf 8y ago
  There are quite a few legitimate entry points. There's a huge body of artists who had no idea about the physiology of human vision but absolutely understood hue, chroma, brightness. They clearly understood the real world was HDR even if they didn't use that term. They understood the massive loss of dynamic range of their artwork, and yet they developed the concept of rendering in order to compress shadow and highlight detail to preserve some sense of the real world when important and to clip it when it wasn't important. And you don't even have to go back 600 years for such examples, even today's amateur photographer learns about these problems and work arounds, without the benefit of learning any physiology. We can't all dedicate the time to read Wyszecki and Stiles as the entry point into color and imaging, albeit of course it's quite useful for those who need to understand such origin stories and the math for all of it.
cmurf 8y ago
It's quite a decent article. I prefer the term "color image encoding" for sRGB IEC61966-2, or ITU-R BT.2020 or DCI-P3, because it encapsulates the primaries (red, green, blue in relation to CIE XYZ), tone response curve (often defined with a gamma function, or the sRGB curve which approximates a curve defined by a gamma function of 2.2 but in fact they aren't the same as the article points out), precision or bit depth, the reference viewing condition, and the reference medium. Very often ignoring the differences in reference viewing condition or medium, gets people into trouble when they do transforms between these encodings and think the environment or medium doesn't have to change. For example there are four variants of DCI-P3, with the same primaries, different white points, and different tone response curves, to account for their different reference viewing conditions. Are they all the same color space? Errr, maybe, yes? Are they all the same color imagine encoding? Definitely not. Same for ITU-R BT.709 vs sRGB; same for Adobe RGB (1998) vs opRGB. As the reference mediums differ, so will the dynamic range. CIE XYZ, based on the 1931 standard observer, continues to be put through its paces. There's also the 1964 standard observer derived from that and supplemental observations and research. There's dozens of color appearance models that have appeared, and continue to be an active area of research. Some try to account for various features of human vision including optical illusions (illusions expose them as if they are a kind of trick or failure, but in fact they're a feature) like simultaneous contrast, Bezold effect, distinguishing between saturation, chroma, and colorfulness, and many other aspects of appearance. Recently there's some understanding the idea of a single standard observer is probably wrong, and how to go about categorizing and handling multiple standard observers (i.e. normal color vision) and that is also yet another area of active research. For anyone interested in color science, or having substantial math and/or computing skill, and looking for unique real world application, I highly recommend IS&T's annual Color Imaging Conference. imaging.org
edwintorok 8y ago
There are also color appearance models, the latest one is CAM16UCS[1], which is simpler than the older CIECAM02[2] model, but harder to find papers explaining it. The Jab[2] representation is more perceptually uniform than Lab. [1] https://colour.readthedocs.io/en/develop/_modules/colour/appearance/cam16.html https://colour.readthedocs.io/en/develop/_modules/colour/app... [2] https://en.wikipedia.org/wiki/CIECAM02 https://en.wikipedia.org/wiki/CIECAM02
chewxy 8y ago
Out of curiosity, what is stopping screen makers from making the entirety of the colour space as defined by the CIE chromaticity diagram its gamut? Is there some fundamental limitation of liquid crystals? I understand for example, CRTs require red phosphor made of yttrium and europium, which somewhat limits the emitted freq. What about LEDs and LCDs? What are the limitations for making a wider gamut? Do note my knowledge in this side of things is close to nil (the specific examples of yttrium and europium as red phosphor was a story told to me by a physicist friend, and it stuck)
- Majromax 8y ago
  > Out of curiosity, what is stopping screen makers from making the entirety of the colour space as defined by the CIE chromaticity diagram its gamut? A display can't support the entire space because of the way colours combine. Any light source acts as a point in the colour space, so a 3-colour display can only cover a triangle. That said, wide gamut monitors do exist. They're of limited utility for simply consuming media; because of the chicken-and-egg effect everything produced for mass consumption already targets the reduced sRGB space or an equivalent. However, a wide-gamut display is useful for content generation, especially when targeting print processes that can use spot colours.
- jacobolus 8y ago
  You can make a much wider gamut display (while still avoiding severe observer metamerism) if you have more primaries. You could even do proper multispectral imaging. It would be awesome if display manufacturers would try to make a display with e.g. 7 primaries in hexagonal pixels. Pretty well every image that currently shows up on smartphone screens already has to be resampled, so there shouldn’t be any inherent reason why you couldn’t have more sophisticated displays requiring slightly more signal processing to target. It would take a bunch of engineering effort, but nothing extraordinary or out of reach of current technology. The big problem is that every other stage of the image processing pipeline from cameras and GPUs to software is built on an existing square-grid-of-RGB-pixels model.
- seandougall 8y ago
  If you take a look at the xyY diagram: > https://ciechanow.ski/images/color_gamut_srgb.svg https://ciechanow.ski/images/color_gamut_srgb.svg the gamut is defined by the triangle connecting the RGB primaries. If you could add a primary at a more pure green, you'd get a bigger triangle and a wider gamut, but still not the entirety of the chromaticity space. If you start adding other components that lie outside that triangle, you start expanding the polygon, but there will always be parts of the diagram it doesn't cover. In order to encompass the entire chromaticity space, you'd have to have a large (technically infinite) number of components. You can never have a component outside the diagram, because the curved hull is the pure chromaticity of light at each wavelength; light outside the curve doesn't exist, and light below the straight "line of purples" is invisible (UV/IR). Theoretically you could get much more coverage by adding more components as new types of emitters are developed. This is happening in high-end theatrical lighting, where you can cram a ton of different LEDs into a fixture, but it's not (yet) practical to cram them all into an on-screen pixel. Things get even more mind-bending when you start illuminating objects with light instead of looking at the light directly, because then there's a benefit to adding components even if they don't expand the gamut. Once you have four or more components, you can make many colors an infinite number of ways (metamers), and they might look absolutely indistinguishable when lighting a white wall, but they'll make skin tones look completely different if you made them by combining just RGB versus, say, amber and blue. That's not really related to your question, but I've just started delving into this area in my day job, and it's exciting and trippy. :)
lettergram 8y ago
I've actually used color spaces in a lot of my work, and it can be highly effective to alter the color space to improve performance of computer vision applications (even the ML kind): https://austingwalters.com/edge-detection-in-computer-vision/ https://austingwalters.com/edge-detection-in-computer-vision... https://austingwalters.com/chromatags/ https://austingwalters.com/chromatags/
zeeshan121 8y ago
GET LATEST MONTHLY COUPONS AND PROMO CODES AT kouponkabla.com

Color Spaces

236 points by tylerchr 8y ago

platz 8y ago

gimp needs a LAB mode

pmoriarty 8y ago
It has one.
- platz 8y ago
  i see. it's just operates very differently; i was hoping to view the result of applying LAB curves in real time on the original image, instead of a multi-step process of decomposing and re-composing each time to view the results. I guess I can just use two windows though

teddyh 8y ago

See also the venerable Color FAQ, a link to which is conspicuously absent: http://www.poynton.com/ColorFAQ.html http://www.poynton.com/ColorFAQ.html

pzone 8y ago

The CIE XYZ color space is fundamental to all modern color space processing. Very cool that it was defined in 1931 and has held up more-or-less unchanged to today. Be sure to click and drag on the 3d cubes!

zokier 8y ago
> Very cool that it was defined in 1931 and has held up more-or-less unchanged to today I wonder if the color mixing experiments have been recreated since the 30s, and if there has been any observable shifts in the curves
- jacobolus 8y ago
  There have been a bunch of additional experiments. But under well-lit conditions with typical light sources, for a typical viewer with normal color vision looking directly at an image / colored object, the 1931 functions are close enough that switching isn’t worth the transition cost. The best recent data can be found at http://www.cvrl.org/ http://www.cvrl.org/

mroche 8y ago

This is a very well written article, and boils down the complexity of color science into very digestible bites. As someone who has to deal with color spaces daily (VFX/CG), having something like this when I started would have been incredibly helpful.

Straw 8y ago

No mention of CIELAB? Unfortunate, its a really neat way to get perceptual uniformity, has intuitive coordinates.

kuon 8y ago

For web design, I use HSLuv[1] which is very handy. [1]: http://www.hsluv.org/ http://www.hsluv.org/

misterdoubt 8y ago
That's damn cool.
Exuma 8y ago
Can you explain in laymans terms to another web designer why that is important? Like what problem does it solve?
- kuon 8y ago
  For example, you want to dynamically generate a background and text color based on a hue. With hsluv you can simple do `hsluv(120, 60%, 10%)` for background color and `hsluv(120, 60%, 90%)` for foreground text color, you will have text on background with 80% perceived luminance difference, you can replace the `120` in my example with whatever hue in the `0-360` range and have consistent results. The idea is that the color space is linear (perceived) with saturation and luminance regardless of hue. You can see it clearly if you compare something like http://www.hslpicker.com http://www.hslpicker.com with http://www.hsluv.org/ http://www.hsluv.org/ if you drag the hue slider on the hsl version, the luminosity varies, it does not on the hsluv version.
  - Exuma 8y ago
    Ahh, very interesting. That is cool.

jacobolus 8y ago

Like many resources about color, this one has an output-medium-centric followed by early-20th-century-experiment-design-centric viewpoint, presumably because that was where the author started and what he read about. But however perfect the exposition might otherwise be, that is the wrong way to start a general discussion of the subject, and the result is inevitably misleading and confusing for non-experts. > it is a pragmatic approach Having tried to talk face-to-face to many people from different backgrounds about color, in practice this doesn’t work. People are starting with too many misconceptions and gaps in foundational knowledge. If you want to understand color you have to start with at least the basics about the physiology of human vision, along with some basic optics. It is all but impossible to properly understand what is going on with RGB displays, understand what “white point” means, etc. without that background. One of the resource recommended at the end, Bruce MacEvoy’s site http://handprint.com/LS/CVS/color.html http://handprint.com/LS/CVS/color.html is a much better starting place, albeit pretty long for someone just interested in the basics. Or there are many books about human color vision and color reproduction, some friendlier for laypeople and others highly detailed and technical. To understand recent color models Mark Fairchild’s book Color Appearance Models is pretty good, and the first half is a pretty accessible general overview. Or for more about color reproduction try Billmeyer and Saltzman’s Principles of Color Technology (recent editions rewritten by Roy Berns). For more about colorimetry try Hunt & Pointer’s book Measuring Color. Or for something historical, try Kuehni’s Color Space and Its Divisions.

cmurf 8y ago
There are quite a few legitimate entry points. There's a huge body of artists who had no idea about the physiology of human vision but absolutely understood hue, chroma, brightness. They clearly understood the real world was HDR even if they didn't use that term. They understood the massive loss of dynamic range of their artwork, and yet they developed the concept of rendering in order to compress shadow and highlight detail to preserve some sense of the real world when important and to clip it when it wasn't important. And you don't even have to go back 600 years for such examples, even today's amateur photographer learns about these problems and work arounds, without the benefit of learning any physiology. We can't all dedicate the time to read Wyszecki and Stiles as the entry point into color and imaging, albeit of course it's quite useful for those who need to understand such origin stories and the math for all of it.

cmurf 8y ago

It's quite a decent article. I prefer the term "color image encoding" for sRGB IEC61966-2, or ITU-R BT.2020 or DCI-P3, because it encapsulates the primaries (red, green, blue in relation to CIE XYZ), tone response curve (often defined with a gamma function, or the sRGB curve which approximates a curve defined by a gamma function of 2.2 but in fact they aren't the same as the article points out), precision or bit depth, the reference viewing condition, and the reference medium. Very often ignoring the differences in reference viewing condition or medium, gets people into trouble when they do transforms between these encodings and think the environment or medium doesn't have to change. For example there are four variants of DCI-P3, with the same primaries, different white points, and different tone response curves, to account for their different reference viewing conditions. Are they all the same color space? Errr, maybe, yes? Are they all the same color imagine encoding? Definitely not. Same for ITU-R BT.709 vs sRGB; same for Adobe RGB (1998) vs opRGB. As the reference mediums differ, so will the dynamic range. CIE XYZ, based on the 1931 standard observer, continues to be put through its paces. There's also the 1964 standard observer derived from that and supplemental observations and research. There's dozens of color appearance models that have appeared, and continue to be an active area of research. Some try to account for various features of human vision including optical illusions (illusions expose them as if they are a kind of trick or failure, but in fact they're a feature) like simultaneous contrast, Bezold effect, distinguishing between saturation, chroma, and colorfulness, and many other aspects of appearance. Recently there's some understanding the idea of a single standard observer is probably wrong, and how to go about categorizing and handling multiple standard observers (i.e. normal color vision) and that is also yet another area of active research. For anyone interested in color science, or having substantial math and/or computing skill, and looking for unique real world application, I highly recommend IS&T's annual Color Imaging Conference. imaging.org

edwintorok 8y ago

There are also color appearance models, the latest one is CAM16UCS[1], which is simpler than the older CIECAM02[2] model, but harder to find papers explaining it. The Jab[2] representation is more perceptually uniform than Lab. [1] https://colour.readthedocs.io/en/develop/_modules/colour/appearance/cam16.html https://colour.readthedocs.io/en/develop/_modules/colour/app... [2] https://en.wikipedia.org/wiki/CIECAM02 https://en.wikipedia.org/wiki/CIECAM02

chewxy 8y ago

Out of curiosity, what is stopping screen makers from making the entirety of the colour space as defined by the CIE chromaticity diagram its gamut? Is there some fundamental limitation of liquid crystals? I understand for example, CRTs require red phosphor made of yttrium and europium, which somewhat limits the emitted freq. What about LEDs and LCDs? What are the limitations for making a wider gamut? Do note my knowledge in this side of things is close to nil (the specific examples of yttrium and europium as red phosphor was a story told to me by a physicist friend, and it stuck)

Majromax 8y ago
> Out of curiosity, what is stopping screen makers from making the entirety of the colour space as defined by the CIE chromaticity diagram its gamut? A display can't support the entire space because of the way colours combine. Any light source acts as a point in the colour space, so a 3-colour display can only cover a triangle. That said, wide gamut monitors do exist. They're of limited utility for simply consuming media; because of the chicken-and-egg effect everything produced for mass consumption already targets the reduced sRGB space or an equivalent. However, a wide-gamut display is useful for content generation, especially when targeting print processes that can use spot colours.
jacobolus 8y ago
You can make a much wider gamut display (while still avoiding severe observer metamerism) if you have more primaries. You could even do proper multispectral imaging. It would be awesome if display manufacturers would try to make a display with e.g. 7 primaries in hexagonal pixels. Pretty well every image that currently shows up on smartphone screens already has to be resampled, so there shouldn’t be any inherent reason why you couldn’t have more sophisticated displays requiring slightly more signal processing to target. It would take a bunch of engineering effort, but nothing extraordinary or out of reach of current technology. The big problem is that every other stage of the image processing pipeline from cameras and GPUs to software is built on an existing square-grid-of-RGB-pixels model.
seandougall 8y ago
If you take a look at the xyY diagram: > https://ciechanow.ski/images/color_gamut_srgb.svg https://ciechanow.ski/images/color_gamut_srgb.svg the gamut is defined by the triangle connecting the RGB primaries. If you could add a primary at a more pure green, you'd get a bigger triangle and a wider gamut, but still not the entirety of the chromaticity space. If you start adding other components that lie outside that triangle, you start expanding the polygon, but there will always be parts of the diagram it doesn't cover. In order to encompass the entire chromaticity space, you'd have to have a large (technically infinite) number of components. You can never have a component outside the diagram, because the curved hull is the pure chromaticity of light at each wavelength; light outside the curve doesn't exist, and light below the straight "line of purples" is invisible (UV/IR). Theoretically you could get much more coverage by adding more components as new types of emitters are developed. This is happening in high-end theatrical lighting, where you can cram a ton of different LEDs into a fixture, but it's not (yet) practical to cram them all into an on-screen pixel. Things get even more mind-bending when you start illuminating objects with light instead of looking at the light directly, because then there's a benefit to adding components even if they don't expand the gamut. Once you have four or more components, you can make many colors an infinite number of ways (metamers), and they might look absolutely indistinguishable when lighting a white wall, but they'll make skin tones look completely different if you made them by combining just RGB versus, say, amber and blue. That's not really related to your question, but I've just started delving into this area in my day job, and it's exciting and trippy. :)

lettergram 8y ago

I've actually used color spaces in a lot of my work, and it can be highly effective to alter the color space to improve performance of computer vision applications (even the ML kind): https://austingwalters.com/edge-detection-in-computer-vision/ https://austingwalters.com/edge-detection-in-computer-vision... https://austingwalters.com/chromatags/ https://austingwalters.com/chromatags/

zeeshan121 8y ago

GET LATEST MONTHLY COUPONS AND PROMO CODES AT kouponkabla.com