4 ms·

Computer colors aren’t consistent

29 points by hodgesmr 11y ago

TD-Linux 11y ago
Color theory has been extremely well researched for quite a long time. If you are not familiar, I would suggest reading up on the CIE 1931 color space [1]. Some of the confusion in the article relates to terminology. >When we say “yellow” we typically mean a pure monochromatic light with a wavelength of about 570nm. This is not actually what most people mean. Colors like "yellow" usually mean the tristimulus values that produce the visible color yellow. This is a totally different concept than a single-wavelength light source. This becomes more important for colors like "purple". [1] https://en.wikipedia.org/wiki/CIE_1931_color_space https://en.wikipedia.org/wiki/CIE_1931_color_space
- kuschku 11y ago
  But CIE color space still doesn’t help with dichromatic people
yellowapple 11y ago
This is why CRT monitors remained popular for as long as they did even with the advent of LCD technology; they tend to be more accurate color-wise, so graphic designers would prefer them.
Taniwha 11y ago
I spent a lot of time working on color matching back in the early 90s when the first cheap Macs large screens came out - back thyen we thought some simple transform could make it all the same - one thing we eventually realised was that there's only so much you can do here - there are no standard human eyes, they vary as much as screens (for example 10% of the male poulation is red-green colorblind)
jacobolus 11y ago
Ideally, each person should probably characterize their display by eye (instead of a hardware device; and for this to work properly we’d need a better user interface than the existing ones), separately for each context it will be looked at in (i.e. a laptop would need several profiles for various lighting conditions). But we can actually get pretty good results for most people, most of the time, by sticking to industry standards. If a display can get the chromaticity of its primaries close to the sRGB spec and bright enough for the context, and content authors design for sRGB, then most people (minus the substantial percentage who have color vision deficiencies) will see something reasonably close to what the content author intended. If the display doesn’t match the spec (like the iPad mini retina display which trades color fidelity for power efficiency), then it would be nice if software would compensate a bit. Unfortunately, mobile operating systems and software don’t ever bother with color management, because it was too computationally expensive for the first 2 generations of iPhones. It’s sad that this is still true for iOS, considering how Apple pioneered color management software in the 90s. For folks with color vision deficiencies, we could probably do better than current displays, but there are several types of color vision deficiencies so it’s going to be pretty hard to satisfy everyone.
- anon4 11y ago
  >sRGB If we're going to be fixing colour, we can start with the abomination that is sRGB. It has a nontrivial RGB<->sRGB conversion function and you must convert your images to/from it for every single operation like alpha blending (i.e. font rendering with anti-aliasing) and scaling, or the results will be wrong. Sometimes only a bit wrong, sometimes spectacularly wrong. We should just switch to 16 bits per channel linear RGB. People just get everything else so wrong.
  - jacobolus 11y ago
    My point is about getting the display primaries close to the industry-standard spec. Whether you gamma-encode your image data or not isn’t really relevant to that. For saving images to disk or transmitting them over the wire (e.g. images making up part of a web page), it’s a great advantage in most cases to use gamma-encoded rather than linear values, because we want to save bandwidth and storage. For operations internal to e.g. an image editor, just use floating point numbers and a linear space.
  - zackmorris 11y ago
    Totally agree. RGB<->sRGB conversions are a huge mess, especially on young platforms like iOS. 16 bit channels would be easy for everyone to understand, and would eliminate pseudoscience jargon like high dynamic range (HDR).
datenwolf 11y ago
Good morning… another one who woke up… finally. Anyone who was active in computer graphics in the past century knows about this. This is why there are contact color spaces like CIE XYZ or CIE Lab. Yet for some reason we're still using desktop compositors that work in imprecise and small gamut sRGB instead of a contact color space.
jlebrech 11y ago
that's why people who work with graphics use a calibrator.

29 points by hodgesmr 11y ago

Color theory has been extremely well researched for quite a long time. If you are not familiar, I would suggest reading up on the CIE 1931 color space [1]. Some of the confusion in the article relates to terminology. >When we say “yellow” we typically mean a pure monochromatic light with a wavelength of about 570nm. This is not actually what most people mean. Colors like "yellow" usually mean the tristimulus values that produce the visible color yellow. This is a totally different concept than a single-wavelength light source. This becomes more important for colors like "purple". [1] https://en.wikipedia.org/wiki/CIE_1931_color_space https://en.wikipedia.org/wiki/CIE_1931_color_space

kuschku 11y ago
But CIE color space still doesn’t help with dichromatic people

yellowapple 11y ago

This is why CRT monitors remained popular for as long as they did even with the advent of LCD technology; they tend to be more accurate color-wise, so graphic designers would prefer them.

Taniwha 11y ago

I spent a lot of time working on color matching back in the early 90s when the first cheap Macs large screens came out - back thyen we thought some simple transform could make it all the same - one thing we eventually realised was that there's only so much you can do here - there are no standard human eyes, they vary as much as screens (for example 10% of the male poulation is red-green colorblind)

jacobolus 11y ago

Ideally, each person should probably characterize their display by eye (instead of a hardware device; and for this to work properly we’d need a better user interface than the existing ones), separately for each context it will be looked at in (i.e. a laptop would need several profiles for various lighting conditions). But we can actually get pretty good results for most people, most of the time, by sticking to industry standards. If a display can get the chromaticity of its primaries close to the sRGB spec and bright enough for the context, and content authors design for sRGB, then most people (minus the substantial percentage who have color vision deficiencies) will see something reasonably close to what the content author intended. If the display doesn’t match the spec (like the iPad mini retina display which trades color fidelity for power efficiency), then it would be nice if software would compensate a bit. Unfortunately, mobile operating systems and software don’t ever bother with color management, because it was too computationally expensive for the first 2 generations of iPhones. It’s sad that this is still true for iOS, considering how Apple pioneered color management software in the 90s. For folks with color vision deficiencies, we could probably do better than current displays, but there are several types of color vision deficiencies so it’s going to be pretty hard to satisfy everyone.

anon4 11y ago
>sRGB If we're going to be fixing colour, we can start with the abomination that is sRGB. It has a nontrivial RGB<->sRGB conversion function and you must convert your images to/from it for every single operation like alpha blending (i.e. font rendering with anti-aliasing) and scaling, or the results will be wrong. Sometimes only a bit wrong, sometimes spectacularly wrong. We should just switch to 16 bits per channel linear RGB. People just get everything else so wrong.
- jacobolus 11y ago
  My point is about getting the display primaries close to the industry-standard spec. Whether you gamma-encode your image data or not isn’t really relevant to that. For saving images to disk or transmitting them over the wire (e.g. images making up part of a web page), it’s a great advantage in most cases to use gamma-encoded rather than linear values, because we want to save bandwidth and storage. For operations internal to e.g. an image editor, just use floating point numbers and a linear space.
- zackmorris 11y ago
  Totally agree. RGB<->sRGB conversions are a huge mess, especially on young platforms like iOS. 16 bit channels would be easy for everyone to understand, and would eliminate pseudoscience jargon like high dynamic range (HDR).

datenwolf 11y ago

Good morning… another one who woke up… finally. Anyone who was active in computer graphics in the past century knows about this. This is why there are contact color spaces like CIE XYZ or CIE Lab. Yet for some reason we're still using desktop compositors that work in imprecise and small gamut sRGB instead of a contact color space.

jlebrech 11y ago

that's why people who work with graphics use a calibrator.