What is a color space?(makingsoftware.com)
makingsoftware.com
What is a color space?
https://www.makingsoftware.com/chapters/color-spaces-models-and-gamuts
51 comments
> What was shocking to me was just how small the gamut of a printer typically is when seen alongside that of a decent display.
What was a nightmare for me when I worked in prepress was how hard it is to get a convincing purple out of a $20K printer. I used tricks which basically produced nothing like purple in a way that gave customers a good purple impression because the things with a closer actual resemblance to purple always looked awful.
What was a nightmare for me when I worked in prepress was how hard it is to get a convincing purple out of a $20K printer. I used tricks which basically produced nothing like purple in a way that gave customers a good purple impression because the things with a closer actual resemblance to purple always looked awful.
Purple and orange were/are common spot colours for that reason.
My father used to work on all sorts of R&D involving things like how much K to use in substitution of CMY without getting desaturated etc. It's a real rabbit hole, especially if you want to reduce the amount of ink used to prevent soaking the paper.
My father used to work on all sorts of R&D involving things like how much K to use in substitution of CMY without getting desaturated etc. It's a real rabbit hole, especially if you want to reduce the amount of ink used to prevent soaking the paper.
One of the tricks we used in situations that allowed it was to get paper that matched the most important color, this has other downsides but they are more manageable. They were always the last run before everything was taken apart and cleaned for maitenance.
That reminded me of a time when a printer manufacturer approached my old team with this problem. They needed a custom driver for a certain region of the world. In this region in a certain industry they liked the highly saturated 'bad' colours from a competitor, and wanted theirs to match. Much paper and ink was spent on this.
There's a lot to cover. HDR is coming to the web (and already exists on native) and there's certainly lots of issues doing it correctly, and, learning all of the various parts and how to deal with them. (HDR input data, HDR processing, HDR output, the display itself which may or may not be HDR and even if it is HDR might only have so much "headroom", etc..., and there are tradeoffs at each step.
Maybe not appropriate for that particular article but definitely appropriate for the site
Even native still has tons of issues, like the fact that AFAICT, no OS does HDR screen capture. You're viewing an HDR image, you ask the OS to capture the screen. It gives you an SDR capture :( On Mac and iOS that's certainly true. On Windows, the XBox Game Bar thingy will actually capture HDR but the OS level PrintScreen method will not, and the popular ShareX will not either.
Maybe not appropriate for that particular article but definitely appropriate for the site
Even native still has tons of issues, like the fact that AFAICT, no OS does HDR screen capture. You're viewing an HDR image, you ask the OS to capture the screen. It gives you an SDR capture :( On Mac and iOS that's certainly true. On Windows, the XBox Game Bar thingy will actually capture HDR but the OS level PrintScreen method will not, and the popular ShareX will not either.
With a few hacks, you can already get web content rendered with HDR: https://fff.kidi.ng/.
If I'm not mistaken, the latest iOS 26 release supports HDR screenshots.
I suspect that the biggest limitation in printing vs. emissive displays is the simple fact that your contrast ratio and color reproduction is severely limited in printing, because the dye is modifying ambient illumination.
This affects brightness and contrast: For emissive displays, you can have emissive values that are several to many orders of magnitude brighter than the 'black point', and more importantly, the primaries are defined by the display, not by ambient illumination.
Part of the magic of HDR displays is manipulating local masking (a human perceptual quirk) to drive bright regions on a display much brighter than the darker regions, so you can achieve even higher contrast ratios than the base technology could achieve (LED back-illuminated LCD panels, for many consumer TVs). Basically, a bright pixel will cause other nearby pixels to be brighter, because you can't see the dark details near a bright region anyway — but other regions could be darker, where you can perceive more detail in the blacks. This is achieved by illuminating sections of the display at significantly higher or lower levels, based on what your eyes/brain can actually perceive. That leads to significantly higher contrast ratios.
(As a heuristic: photographers generally say you can only get ~5 stops of contrast out of a print. (That is, bright areas are 2^5 times brighter than the darkest regions.) Modern HDR displays can do 2^10 or better. YMMV.)
But this also affects color... much of the complexity in getting printers to match derives from the interaction between the imperfect gamut caused by differing primaries, as filtered through human perception (and/or perceptual models). But you can't control the ambient illumination, so you're at the mercy of whatever the spectrum of your illumination is, plus whatever adaptation the viewer has. This feels fundamentally impossible to do "correctly" under all circumstances.
Which is to say, the original sin of color theory is the dimensional collapse from a continuous spectrum to a 3-dimensional, discretized representation. It's a miracle we can see color at all...!
This affects brightness and contrast: For emissive displays, you can have emissive values that are several to many orders of magnitude brighter than the 'black point', and more importantly, the primaries are defined by the display, not by ambient illumination.
Part of the magic of HDR displays is manipulating local masking (a human perceptual quirk) to drive bright regions on a display much brighter than the darker regions, so you can achieve even higher contrast ratios than the base technology could achieve (LED back-illuminated LCD panels, for many consumer TVs). Basically, a bright pixel will cause other nearby pixels to be brighter, because you can't see the dark details near a bright region anyway — but other regions could be darker, where you can perceive more detail in the blacks. This is achieved by illuminating sections of the display at significantly higher or lower levels, based on what your eyes/brain can actually perceive. That leads to significantly higher contrast ratios.
(As a heuristic: photographers generally say you can only get ~5 stops of contrast out of a print. (That is, bright areas are 2^5 times brighter than the darkest regions.) Modern HDR displays can do 2^10 or better. YMMV.)
But this also affects color... much of the complexity in getting printers to match derives from the interaction between the imperfect gamut caused by differing primaries, as filtered through human perception (and/or perceptual models). But you can't control the ambient illumination, so you're at the mercy of whatever the spectrum of your illumination is, plus whatever adaptation the viewer has. This feels fundamentally impossible to do "correctly" under all circumstances.
Which is to say, the original sin of color theory is the dimensional collapse from a continuous spectrum to a 3-dimensional, discretized representation. It's a miracle we can see color at all...!
> the primaries are defined by the display, not by ambient illumination
In itself that is correct, but as you've noted, our own vision system isn't operating like that. The same display brightness and colors will be perceived very differently depending on the ambient light's brightness and color, and can also mean a severe breakdown in the dynamic range that can be made visible via a display.
And this ambient light also clearly impacts how prints are seen.
In itself that is correct, but as you've noted, our own vision system isn't operating like that. The same display brightness and colors will be perceived very differently depending on the ambient light's brightness and color, and can also mean a severe breakdown in the dynamic range that can be made visible via a display.
And this ambient light also clearly impacts how prints are seen.
Not all people see colors the same. The section on "What is color?" could probably mention color blindness.
This app is one that I rely on for my testing: https://michelf.ca/projects/sim-daltonism/
Add to that the idea that some cultures have fewer color terms than others. This actually impacts upon their perception of color.
Check out "Basic Color Terms: Their Universality and Evolution" by Brent Berlin and Paul Kay.
Check out "Basic Color Terms: Their Universality and Evolution" by Brent Berlin and Paul Kay.
It does not affect perception. This is one of those early anthro/cogsci results that said more about the authors' cultural bias than it did about the people being studied, up there with "Eskimos have a thousand words for snow".
It affects communication. People can still discern the difference between colors, they just don't have an easy way to communicate this difference to others.
The Japanese language until relatively recently didn't have a clear verbal distinction between what we call green and blue in English. That doesn't mean Japanese people can't tell the difference between green and blue. It just means that there is a kind of "blue" that is the sky and a kind of "blue" that is for traffic control lights, and in context nobody is confused.
The same issue can occur within a language between people with differing levels of study of color. A graphic designer might say a particular shade of green is "chartreuse" that his boss instead might call "yellowish green".
It affects communication. People can still discern the difference between colors, they just don't have an easy way to communicate this difference to others.
The Japanese language until relatively recently didn't have a clear verbal distinction between what we call green and blue in English. That doesn't mean Japanese people can't tell the difference between green and blue. It just means that there is a kind of "blue" that is the sky and a kind of "blue" that is for traffic control lights, and in context nobody is confused.
The same issue can occur within a language between people with differing levels of study of color. A graphic designer might say a particular shade of green is "chartreuse" that his boss instead might call "yellowish green".
> up there with "Eskimos have a thousand words for snow".
Bad example. It seems that there is still a lot of disagreement on this matter, much of it rooted in what constitutes a word.
A well-balanced article on the subject here:
https://linguisticdiscovery.com/posts/inuit-words-for-snow/
Edit…
Also, from that article, a fascinating study: ‘Russian blues reveal effects of language on color discrimination’
https://www.pnas.org/doi/full/10.1073/pnas.0701644104
Bad example. It seems that there is still a lot of disagreement on this matter, much of it rooted in what constitutes a word.
A well-balanced article on the subject here:
https://linguisticdiscovery.com/posts/inuit-words-for-snow/
Edit…
Also, from that article, a fascinating study: ‘Russian blues reveal effects of language on color discrimination’
https://www.pnas.org/doi/full/10.1073/pnas.0701644104
This article says almost the exact same things I said in my post. I also don't see where it definitely says the Inuit language has a richer vocabulary for snow than other languages. It just ends with a joke about how such a thing might come to pass. A casual observer here who doesn't bother reading the link might take it as a refutation from your wording, but it actually very strongly supports what I had to say.
As for the "Russian blue" study, I find it strange that the article is so skeptical of unreplicated results in linguistics and yet seems to accept the "Russian blue" study uncritically. I can see at least one glaring flaw: all of the Russian speakers were bi-lingual with English, with at least some of them being so since they were young children. They also discarded 16% of their test data because they deemed responses "too slow", with this discarding more heavily weighted towards the Russian speakers.
As for the "Russian blue" study, I find it strange that the article is so skeptical of unreplicated results in linguistics and yet seems to accept the "Russian blue" study uncritically. I can see at least one glaring flaw: all of the Russian speakers were bi-lingual with English, with at least some of them being so since they were young children. They also discarded 16% of their test data because they deemed responses "too slow", with this discarding more heavily weighted towards the Russian speakers.
Wonder how this applies to animals since their color discrimination would not be impacted by language. Did humans evolve linguistic abilities that alter sensory processing? Seems odd that animals would be able to discriminate colors their eyes can see just fine, but humans would need words to do so.
Good question. I'm no expert, but I guess that the key issue is one of categorization. Without language, it is impossible to effectively categorize our perceptual domain.
It is also true that among mammals, chromatic vision is pretty much restricted to primates. The ability to perceive difference of light is a must if you don't want to become someone else's food. In contrast, chromatic vision is an 'extra' that in many ways serves as a (literally) florid extension to our lives. To me it is no surprise that range of emotion and range of hue are so often associated with each other. Interestingly enough, they are similarly mapped: as a set of differences, rather than as a degree of intensity.
It is also true that among mammals, chromatic vision is pretty much restricted to primates. The ability to perceive difference of light is a must if you don't want to become someone else's food. In contrast, chromatic vision is an 'extra' that in many ways serves as a (literally) florid extension to our lives. To me it is no surprise that range of emotion and range of hue are so often associated with each other. Interestingly enough, they are similarly mapped: as a set of differences, rather than as a degree of intensity.
> perhaps printing is in fact esoteric
You'd be surprised. It may be heading that way, to arrive in a couple more decades, but not yet. It's still quite the industry (that I'm in).
You'd be surprised. It may be heading that way, to arrive in a couple more decades, but not yet. It's still quite the industry (that I'm in).
Nice article, I came across very cool Python library recently too re. colour science - https://www.colour-science.org/
Just started playing with it with my spectrometer based on one of the examples they have, to convert spectral data to a single RGB value.
Just started playing with it with my spectrometer based on one of the examples they have, to convert spectral data to a single RGB value.
Awesome! Maintainer here :) We have a companion repo that might be of interest to you: https://github.com/colour-science/colour-specio
Cool, thanks for the heads up!
Really great introductory article on color space in general, I really appreciate that they touched on perceptual uniformity and how we all perceive colors differently. It’s great to find out that applications like Oklab came out recently to fix this by manipulating the distances between colors to try compensate for human perception while also being more straightforward to calculate so that it can be used in real-time applications. Also the UI of this blog post was so aesthetically pleasing, that it was worth burning my retinas with the light mode
The article skipped a bit on explaining the CIE-XYZ space. Although it is quite simple.
There is one axis Grayscale - that gives you the Brightess value. You can use that for black/white TV.
Now for colour you have 2 dimensions left. For these you pick one axis, where the eye is most sensitive - and perpendicular to that you have one axis, where the eye is least sensitive.
This colour space is the oldest and all axis there make sense. You can easily compare brightness, and you e.g. can assign more bandwidth to the sensitive colour axis if you want.
There is one axis Grayscale - that gives you the Brightess value. You can use that for black/white TV.
Now for colour you have 2 dimensions left. For these you pick one axis, where the eye is most sensitive - and perpendicular to that you have one axis, where the eye is least sensitive.
This colour space is the oldest and all axis there make sense. You can easily compare brightness, and you e.g. can assign more bandwidth to the sensitive colour axis if you want.
That's a terrific article!
I spent the majority of my career at an optical equipment manufacturer, and we wrote a color management system to handle 48-bit color, before any OS manufacturers had it.
Non-trivial stuff, but powerful.
I know of at least one technology that works by converting to an esoteric color space, messing with image data there, then converting back.
I spent the majority of my career at an optical equipment manufacturer, and we wrote a color management system to handle 48-bit color, before any OS manufacturers had it.
Non-trivial stuff, but powerful.
I know of at least one technology that works by converting to an esoteric color space, messing with image data there, then converting back.
Indeed... and it was always a huge amount of fun to find that converting from one space to another was not always reversible without distortion!
This is a great primer. I used to work in this field, but not as a colour scientist.
> So why do we have so many different color spaces?
I think there was a missing piece here. Different representations of colour are useful for different things. I'm not going to give any secrets away but...if your trade involves finding out how close one colour is to another, then something that represents colours as points in space could make the maths easier. Then if you wanted to know if one colour was brighter than another, then something that represents a colour with a separate 'brightness factor' would make that trivial.
> So why do we have so many different color spaces?
I think there was a missing piece here. Different representations of colour are useful for different things. I'm not going to give any secrets away but...if your trade involves finding out how close one colour is to another, then something that represents colours as points in space could make the maths easier. Then if you wanted to know if one colour was brighter than another, then something that represents a colour with a separate 'brightness factor' would make that trivial.
as post production guy living outside US/EU, a decade or so ago i was surprised that colorimeter was supposed to be regularly calibrated by another, more expensive, device.
I think that depends on what you mean by 'supposed to'. I don't recall this being suggested for normal applications, and if you have a very tight specification then a colorimeter may not be the correct instrument in the first place. As I said above, I'm not a colour scientist, but I do know something about what was done in practice in certain industries.
i'm in post-production so lots of color spaces & monitorings. colorimeters drifted with age/usage, so annual check for the first two years or so is good enough but then gradually afterwards it's good practice to calibrate more frequent, and more likely there's a small fee for that. spectrometers (or spectroradiometer can't be sure) are still insanely expensive.
We were big fans of Marti’s LCMS.
I can't remember the technical reason we didn't use it...we mostly rolled our own. We were very lucky to have someone who knew how!
IIRC the licensing was restrictive for commercial use but it's been a long time. I think we ended up throwing him a few grand for a different license, an option I'd never considered before and one that's stuck with me.
One crazy thing you can do with color spaces is dramatically improve detection algorithms and store massive data.
https://austingwalters.com/chromatags/
Think of it this way, a QR code is binary. If you modify color spaces correctly you can get 6 bits (or more) per pixel. In addition, you can improve the detection at distance, localization for robots, and speed (120 fps).
Done this to great effect previously and you can do a lot of awesome things with it. Pretty much the easiest hack in computer vision.
https://austingwalters.com/chromatags/
Think of it this way, a QR code is binary. If you modify color spaces correctly you can get 6 bits (or more) per pixel. In addition, you can improve the detection at distance, localization for robots, and speed (120 fps).
Done this to great effect previously and you can do a lot of awesome things with it. Pretty much the easiest hack in computer vision.
This is a great article about the topic! It covers everything.
But can someone explain this to me?
>Light is technically something called electromagnetic radiation and it has a frequency and wavelength. That wavelength can vary, depending on the energy of the wave. High energy waves have a higher frequency and shorter wavelength, and low energy waves have a lower frequency and longer wavelength.
>This means that the same amount of energy at different wavelengths will not be perceived as the same brightness. For example, a light with a wavelength of 555 nm (green) will appear brighter than a light with a wavelength of 450 nm (blue) even if they have the same energy.
The article asserts that the wavelength (thus color) changes with the amount of energy, but then it says that you can have light of different wavelengths (color) with "the same energy."
But can someone explain this to me?
>Light is technically something called electromagnetic radiation and it has a frequency and wavelength. That wavelength can vary, depending on the energy of the wave. High energy waves have a higher frequency and shorter wavelength, and low energy waves have a lower frequency and longer wavelength.
>This means that the same amount of energy at different wavelengths will not be perceived as the same brightness. For example, a light with a wavelength of 555 nm (green) will appear brighter than a light with a wavelength of 450 nm (blue) even if they have the same energy.
The article asserts that the wavelength (thus color) changes with the amount of energy, but then it says that you can have light of different wavelengths (color) with "the same energy."
He confuses two different places there is energy. Light is made (this is a little bit of a cheat) of photons. Each photon has a wavelength λ, and a per photon energy E where E = hc/λ, h is Plank's constant and c is the speed of light constant. So energy and wavelength per photon completely determines each other (and the color of the light of that single photon).
These energies are very small. You can add a lot of photons per second, increasing the brightness of that color, and this now has the energy per second of all those photons. So you can have a lot of red photons which sum to some energy, or a different number of blue photons that sums to (very, very close) the total red energy.
These are the two energies he confuses in those two places.
These energies are very small. You can add a lot of photons per second, increasing the brightness of that color, and this now has the energy per second of all those photons. So you can have a lot of red photons which sum to some energy, or a different number of blue photons that sums to (very, very close) the total red energy.
These are the two energies he confuses in those two places.
The amount of energy in a wave depends on both its wavelength and amplitude.
So a "blue wave" has more energy than a "red wave" if both have the same amplitude (blue has a shorter wavelength, and energy is inversely proportional to wavelength). But you can have a "blue wave" with the same energy as a "red wave" if you increase the amplitude of the "red wave" to compensate for its longer wavelength.
So a "blue wave" has more energy than a "red wave" if both have the same amplitude (blue has a shorter wavelength, and energy is inversely proportional to wavelength). But you can have a "blue wave" with the same energy as a "red wave" if you increase the amplitude of the "red wave" to compensate for its longer wavelength.
Oh, that makes sense! Thanks.
So that means blue has more energy because it pulsates faster, and in spite of this we're less sensitive to it than we are to red, which pulsates slower. It's like our light sensitivity forms some sort of bell curve.
So that means blue has more energy because it pulsates faster, and in spite of this we're less sensitive to it than we are to red, which pulsates slower. It's like our light sensitivity forms some sort of bell curve.
I love that this opens with a matter-of-fact statement that colors aren't real, just a phenomenon of perception. This has come up randomly in my life a few times and people always look at me like I'm stoned outta my mind.
And what is not just a phenomenon of perception?
The light hitting our eyes is not a phenomenon of perception. Each photon hitting our eyes has a certain wavelength, but the resulting color we see is a perception.
A related concept is sound vs psychoacoustics[1]. Sound is just pressure waves, but what we hear is a perception and has all sorts of aspects like masking[2]. The pressure waves contains two different signals but thanks to masking we might only perceive one.
Personally I think color constancy[3] really drives home that color is a perception and not something fundamental like the wavelength of photons.
[1]: https://en.wikipedia.org/wiki/Psychoacoustics
[2]: https://en.wikipedia.org/wiki/Auditory_masking
[3]: https://en.wikipedia.org/wiki/Color_constancy
A related concept is sound vs psychoacoustics[1]. Sound is just pressure waves, but what we hear is a perception and has all sorts of aspects like masking[2]. The pressure waves contains two different signals but thanks to masking we might only perceive one.
Personally I think color constancy[3] really drives home that color is a perception and not something fundamental like the wavelength of photons.
[1]: https://en.wikipedia.org/wiki/Psychoacoustics
[2]: https://en.wikipedia.org/wiki/Auditory_masking
[3]: https://en.wikipedia.org/wiki/Color_constancy
The association of specific wavelengths to hue exists only in our perception, nowhere else.
On the other hand, we perceive temperature yet its impact on the physical world is universal.
On the other hand, we perceive temperature yet its impact on the physical world is universal.
Mathematical proofs based upon axioms?
Fun fact: just like people have different heights, they have different perceptions of color!
A very clear article, written in very accessible language. One potentially relevant thing not covered is the existence of super blacks and super whites, values which (as a budding vfx artist) I often need to consider. These are values which exist in the abstract sence, but not in the perceptual sence. Specifically, values darker than black and lighter than white.
They are as good a demonstration as any of the notional nature of color spaces.
They are as good a demonstration as any of the notional nature of color spaces.
Recently saw this useful video on colors https://aeon.co/videos/after-centuries-of-trying-weve-yet-to...
There's a nice YouTube video that covers a bit of the history of colour spaces to I was watching yesterday https://youtu.be/7KYwi2F5Ce4?si=DUxIWDe_SEjwhdeI
Great read and excellent website. Is there a specific name for this web design style? I would have liked to see more websites like this or similar ones for inspiration.
Such an amazing article.
Hope the author continues with the rest of the content in the same manner.
Such a great read
If I had to criticize, I wish it had talked a bit more about printer color profiles (although in this paperless, web-world we live in, perhaps printing is in fact esoteric).
Unlike displays, a printer can't be simply defined with three primaries and a white point. Printer profiles can be quite large as they rely on someone having printed out a copious number of swatches on a given paper type and then measured (with some kind of colorimeter) device independent color values for each swatch. Those are used to build a large table for mapping from device independent color spaces to the printer's gamut.
Those large tables make the profile so large. And then of course interpolation is still required when mapping from a device independent color space to the printer profile. (Now imagine too that you need a different profile for each type of paper you might want to print to since each can represent color differently — plain paper unable to get the levels of saturation that a coated paper can.)
What was shocking to me was just how small the gamut of a printer typically is when seen alongside that of a decent display.
Consider that, in print, you'll never see an image as vivid as you can display on a nice, modern display. (And then consider that there are colors in nature so vivid that even a modern display cannot represent them. Just look at how much color is outside the triangle on the CIE "shark fin" color representation.)
Also not touched on (did I miss it?), all the math presented to map from one color space to another, also allows for "soft proofing" — where in fact you might match to a printer ICC profile but then take the result and match again to the user's display to give you a "preview" of what will be lost when going to said printer.