If the RGB color model is capable of 16.7 million colors, has anyone ever calculated how many colors the average person’s eye is capable of perceiving?
Any idea on how many colors exist in nature?
How large is the gamut of the eye to the number of colors in nature?
How many colors are in the SRGB color space?
Thanks in advance for any help.
Paul
Sorry for the delay in getting back to you on this post. We also created a new forum for color science questions just like this one.
If the RGB color model is capable of 16.7 million colors, has anyone ever >calculated how many colors the average person’s eye is capable of >perceiving?
a good question. The RGB color model, using 8 bits per channel, is capable of creating 16.7 million ‘number combinations’ but not that many colors…
Any idea on how many colors exist in nature?
see below
How large is the gamut of the eye to the number of colors in nature?
see, now I’ve got to get all philosophical. Colors don’t exist in nature. They exist in our visual system. so…
number of colors in nature = number of colors we can see
and
number of colors we can see = number of colors in nature
How many colors are in the SRGB color space?
Now THAT I can give you an answer for.
A few rough numbers for comparison: (all these values are in cubic CIELab)
Gamut of human color vision: 2.2 - 2.3 million colors
Gamut of G4 Powerbook: 518,733
Gamut of Apple HD Cinema Display: 928,189
so, even though each of the displays has 16.7 million addressable RGB values, the Apple HD Cinema Display can display almost 410,000 more colors. A significant difference to say the least. It could be argued that the PowerBook really only needs about 19 bits for its display - who wants to start an argument though?
A few more values for comparison.
sRGB = 909,800
Adobe RGB = 1,320,550
SWOP TR001 = 297,500
ISOcoated from ECI = 400,350
please note, all these values are very rough. They are best used only for broad comparisons.
The number of colours that can be represented on a display are mostly determined by the colors of the colorants (filters for LCD, phosphors for CRTs). The more saturated the colorants the larger the gamut and the more individual colors can be displayed.
Hope this helps,
Steve
So why doesn’t my print mach the monitor again?
Sorry, couldn’t resist.
At 8:04 AM -0700 7/28/05, SonyDADC wrote:
[quote:520ad1254b=“Steve Upton”]
…A few more values for comparison.
sRGB = 909,800
SWOP TR001 = 297,500
[/quote:520ad1254b]
So why doesn’t my print mach the monitor again?
yup, that’s worth a wry chuckle
Actually, A portion of this last post is from a post I did to a mailing list where I was trying to make it clear to a particularly stubborn user that just because a Powerbook display can be addressed with 16.7 million numbers doesn’t mean that it will display all the colors of an inkjet print…
No matter how many numbers, facts and descriptions I supplied I’m not sure he got it… .I’m not sure he wanted to…
Regards,
Steve
o Steve Upton CHROMiX www.chromix.com
o (hueman) 866.CHROMiX
–
Post generated from email list
Steve,
Will ColorThink Pro include a function to enable me to find out the gamut volume in cubic CIELab (or any other suitable unit) of profiles and images?
TIA
Alan
At 4:32 AM -0700 7/29/05, alanrew wrote:
[quote:1a9f1157d4=“Steve Upton”]
A few rough numbers for comparison: (all these values are in cubic CIELab)
sRGB = 909,800
Adobe RGB = 1,320,550
SWOP TR001 = 297,500
ISOcoated from ECI = 400,350
[/quote:1a9f1157d4]
Steve,
Will ColorThink Pro include a function to enable me to find out the gamut volume in cubic CIELab (or any other suitable unit) of profiles and images?
you betcha!
that’s where these numbers are from… although don’t hold me to them. I am tweaking the algorithm for gamut volume calculation (they shouldn’t be massively different though)
Regards,
Steve
o Steve Upton CHROMiX www.chromix.com
o (hueman) 866.CHROMiX
–
Post generated from email list
Any idea how many colors a LaCie CRT monitor is capable of displaying?
I"ve got a class to teach and need a few numbers for comparason.
Thanks for the help.
At 4:52 PM -0700 8/8/05, PaulV wrote:
Any idea how many colors a LaCie CRT monitor is capable of displaying?
I"ve got a class to teach and need a few numbers for comparason.
not off the top of my head, no…
It will be somewhere in between sRGB and Adobe RGB (closer to sRGB, I suspect)
Regards,
Steve
o Steve Upton CHROMiX www.chromix.com
o (hueman) 866.CHROMiX
–
Post generated from email list
Good news!
Once you can do that, maybe you can then start telling me things like
For a long time I’ve been looking for a way to quantify the ‘out-of-gamutness’ of an image. Maybe your volume metrics will help.
Regards
Alan
Wow, I just discovered this forum, and it looks great. (I just had to say that.)
Anyway, Steve, your comment makes me think. How does the numbers you mention relate to quantizations problems with really big versus small profiles. I always thought that the bigger your profile’s gamut, the more “space” between each colour. That is to say, with 8 bits, those 16.7 million number combination have to be spread out across more visual colours. But, if you say that the human eye can only see 2.3 million cubic Lab units, then it sounds like 16.7 million number combinations ought to easily characterize all the colours the eye can see. It sounds like 8 bits is quite excessive, let alone 16 bits (or even 32 bits now).
To put it concretely, I’m always worried that my scanner profiles are missing visual colours because their gamut is so big. I wonder if a profile generated from a 16 bit image would better characterize the device. But by the sounds of what you’re saying, 8 bits will quite easily capture all the perceivable colours. 16.7 million number combinations far exceeds the actual number of cubic Lab units within the gamut.
What do you think? Do you get what I’m saying?
Thanks,
At 4:31 AM -0700 8/24/05, alanrew wrote:
Good news!
Once you can do that, maybe you can then start telling me things like
- the gamut volume of the out of gamut colours in an image, relative to a given output profile;
- the gamut volume of the overlapping region of two profiles or two images or in general two 3d-plottable spaces.
For a long time I’ve been looking for a way to quantify the ‘out-of-gamutness’ of an image. Maybe your volume metrics will help.
yes, well, this is considerably more complicated - but not impossible though.
Basically what you are looking for is what they call boolean intersections (or something like that) in 3D graphics…
It is on my list to look at in future versions… hmmm… yeah I think it could be doable…
Regards,
Steve
o Steve Upton CHROMiX www.chromix.com
o (hueman) 866.CHROMiX
–
Post generated from email list
At 1:39 PM -0700 8/25/05, devlinr wrote:
Wow, I just discover this forum, and it looks great. (I just had to say that.)
Anyway, Steve, your comment makes me think. How does the numbers you mention relate to quantizations problems with really big versus small profiles. I always thought that the bigger your profile’s gamut, the more “space” between each colour. That is to say, with 8 bits, those 16.7 million number combination have to be spread out across more visual colours. But, if you say that the human eye can only see 2.3 million cubic Lab units, then it sounds like 16.7 million number combinations ought to easily characterize all the colours the eye can see. It sounds like 8 bits is quite excessive, let alone 16 bits (or even 32 bits now).
that’s a very good point. It all comes down to WHICH colors those numbers represent. It is surprising to think that in a number space of 16.7 million numbers that we might not be able to actually capture 2.3 million or so colors… but it’s true.
I suppose the first thing that comes to mind is gamma. If you fool around with working spaces in Photoshop and feed 50,0,0 into the color picker you will see what that comes to for your current working space. Gamma 1.8 comes to 100 or so whereas 2.2 comes to 117-18 or so if I remember right. This means that with 1.8, there are about 100 values available to encode shadows and 155 or so for highlights. With 2.2 it is more even with 117 for shadows and… well, you do the math… you know what I mean…
It would seem like 100 should be fine but there are instances where it is not enough.
To put it concretely, I’m always worried that my scanner profiles are missing visual colours because their gamut is so big. I wonder if a profile generated from a 16 bit image would better characterize the device. But by the sounds of what you’re saying, 8 bits will quite easily capture all the perceivable colours. 16.7 million number combinations far exceeds the actual number of cubic Lab units within the gamut.
high bit is typically a good idea in scanners always… well, truthfully it doesn’t necessarily make that much difference when building a profile but it does when scanning so you have some depth if you want to make some significant changes in tone.
The most important thing I have found when profiling a scanner is to set the gamma so that the L=50 patch on the scanner target (look at the reference file that comes w/ the target to see which patches qualify) comes out to about 127 RGB value. The RGB’s won’t be perfectly equal - that’s why you’re profiling after all - but get the gamma set right and your encoding will be good. We also have users build profiles at a higher and lower gamma setting to catch images that fall out of the typical range.
The size of the gamut of a scanner profile is an odd beast at best. If you don’t “hand” profiling software RGB values at their extremes (which you really never do, or should) then it has to extrapolate and that is up to the whim of the software’s designer.
Overall the gamut size isn’t too much of a concern if you scan in 16bit, correct for tone as needed then downsample to 8.
Regards,
Steve
o Steve Upton CHROMiX www.chromix.com
o (hueman) 866.CHROMiX
–
Post generated from email list
Probably the biggest difference in 16 bit profiles/ and or high bit images to build scanner profiles is not so much in that they could create more color or gamut but in that the bit depth allows less quantisisation errors not only when using the profile but also in it’s creation . When the software devs make the algorithms work there is always some type of reference space to which grid points within this space will have correlated output. The higher the precision the better the chance of formulating numbers with higher accuracy.
But this is all theory and pure input images do not exist hence what your formattted scan of a chart will be is less than the ideal of the model>
Next question which one will always ask is which scan profiler then does this the best?
And to Alan, Alwan software does some of the things you thought were neat. OF course when I stop having fun in Colorthink then I’ll look elsewhere. So Steve bring it on!