The CIE 2 deg versus 10 deg models
Timo Autiokari
I have been experimenting with these 1931 (2 degree) and 1964 (10
degree) CIE Standard Observer models. First I did the following:
1) I took 175 spectras, real Spectroscan measured spectras, selected
from a larger group of spectras in such way that the resulting
trichromatic colors will fit to the CRT gamut in both cases (as
converted by the 2 deg model as well as converted by the 10 deg model)
therefore enabling me to evaluate using the CRT.
2) Then convert the spectras by the 2 deg model to XYZ and from that
to nativePC RGB profile and created a patches that had the size of
about 2 degree when my nose is at 30 cm from the screen.
3) Then convert the same spectras by the 10 deg model to XYZ and from
that to nativePC RGB profile and created a patches that have the size
of about 10 degree when my nose is at 30 cm from the screen.
4) Finally arranged the patches side by side and set the background as
gray (tried many different grays)
Amazingly, there was two kind of results:
Either A) there is no visible difference between the patches or B)
there was a visible difference between them. When there was no visible
difference the RGB values were also pretty close between the
comparison pairs.
The expected result naturally is that none of the 175 comparisons
would show a visible difference. But many do.
So after pretty hard pondering I eventually found a (very obvious)
method to test this further:
I made the same experiment as above but now I created 10 degree
patches from the 2 degree model and 2 degree patches from the 10
degree model.
Surprisingly the results were _exactly_ the same as in the previous
experiment: The comparison pairs that had no visible difference in the
previous experiment had no visible difference it the latter experiment
either. And the comparison pairs that had visible difference in the
previous experiment had very similar visible difference it the latter
experiment.
So, at this point it looks to me that the existence of these two
models is absolutely gratuitous and groundless.
Do you think that the above experiment is OK or does it have some kind
of fault? I will write a web page about this with plenty of examples
but there is quite a lot of work involved so I'd like to discuss about
the testing methodology first so that I need not to redo all the
examples.
Timo Autiokari
Douglas G. Cummins
Timo Autiokari wrote:
<snip>
> 2) Then convert the spectras by the 2 deg model to XYZ and from that
> to nativePC RGB profile and created a patches that had the size of
> about 2 degree when my nose is at 30 cm from the screen.
>
> 3) Then convert the same spectras by the 10 deg model to XYZ and from
> that to nativePC RGB profile and created a patches that have the size
> of about 10 degree when my nose is at 30 cm from the screen.
> Do you think that the above experiment is OK or does it have some kind
> of fault? I will write a web page about this with plenty of examples
> but there is quite a lot of work involved so I'd like to discuss about
> the testing methodology first so that I need not to redo all the
> examples.
I'm not confident in your methodology. At 30 cm, you'd create a "patch"
with an effective diameter of 1 cm for a 2° subtense and 5 cm for a 10°
subtense. The two different sizes would make side-by-side comparisons
problematic. Also, the eye sees and interprets the spectra - not an xy
value. You are limited by the gamut of your monitor or your printer.
The two different observation angles are meant for different
applications. The CIE 1931 2° observer is meant for distant or small
color sources (visual field angle subtending no more than 4°) - for
instance, I use CIE 1931 for automotive lighting colors. The CIE 1964
10° observer supplement is meant for near or large color sources (visual
field angle subtending an angle greater than 4°) - which is why it's
used extensively in the textile and graphics industries. You might try
observing the same patch at two different distances and see if both
observation points appear the same color (I don't know you can do that
with any reliability though).
There is an interesting note in ASTM E308 for the CIE 1964 definition:
"Users should be aware that the CIE 1964 (10°) supplementary system and
standard observer assume no contribution or constant contribution of
rods to vision. Under some circumstances, such as in viewing highly
metameric pairs in any but very high light levels (where the rods are
saturated), the amount of rod participation can vary between the members
of the pair. This is not accounted for by a trichromatic system of
colorimetry. The 10° system and observer hould not be used in such
circumstances."
That said, both observers have generally the same results based on
experimental evidence from relatively small samplings. It's entirely
possible that if you were to try repeating both experiments but on a
larger scale/sampling you could come up with results that are a mixture
of the two standards.
--
Douglas Cummins
Calcoast - ITL
Danny Rich
When you attempt to "characterize" the difference between the 1931 and the
1964 you are basically assess the level of observer metamerism. It is very
very difficult to assess metamerism with non-metameric stimuli. The most
common way of "visualizing" the difference between the size of the visual
field on the retina is to make a metameric match for a large field (>10°)
and then move back from the match and see if the match holds. For
non-metameric pairs the match will generally hold (Persistence of Match Law
of Colorimetry). However, for a strongly metameric match the match will
fail at small visual angles (<4°). You might try creating a CMY (inkjet
without K) match to your CRT. Make both patches about the same size and
adjust the brightness of the spot source as reflected by a white paper to be
very similar to the brightness of the monitor. Move up close to the monitor
and adjust the RGB of a CRT patch until it is a very close match to the ink
on paper (near neutrals work best for this as the mixtures of CMY and RGB
will be complex). Then back away from the monitor until the size of the
two patches is less than 4°. See if the CRT patch still matches the printed
match. Generally it does not.
I have seen this demonstrated by the late Dr. Henry Hemmendinger, inventor
of the Davidson & Hemmendinger Color Rule. He built a version of the rule
that was about 2 meters in length. The patches were then about 20 cm wide x
10 cm high and the two patches place one above the other were about 20 cm
square. The rule has a series of colors that range from red to green on the
top series and from blue to green on the bottom series. One pair of patches
could be selected and identified as the "best" match possible under a given
viewing condition. The rule was placed upright on a desk or blackboard and
the observer backed away until the patch size image is less than 3° and the
match would be broken.
Danny Rich
"Timo Autiokari" <timo.au...@aim-dtp.net> wrote in message
news:2kieh193vh32ujhhk...@4ax.com...
bob
experiment in the way you describe it, your visual system has do
develop a sort of "split personality" in a sense that the same eye
needs to evaluate a large and a small patch simultaneously. You still
can make a valid experiment if you make sure that:
1. you view each patch with separate eye, e.g. you have a physical
separation between the patches so it limits the view of each eye to the
corresponding half-view. And note that there can be possible difference
in how your both eyes see colour...
2. You take into account that when you view 10deg field you ignore the
central part of the field and evaluate the periphery only (while the
gaze is fixated on the centre)
And one more point as for the idea of the experiment. I think it is
important to understand that the CIE (or any other possible) average
observers will inevitably fail when tested in conditions which are
different from the ones they were measured in. The way these observers
are validated is not by whether they stand the test of any particular
application, but whether they provide a reasonably good results when
used with whole variety of applications in the many industries they are
used in.
This is to say that you cannot possibly derive a conclusion of a kind
"...existence of these two models is absolutely gratuitous and
groundless" on a basis of your experiment, however carefully
designed.
Erik Nikkanen
> However, for a strongly metameric match the match will
> fail at small visual angles (<4°).
Hi Danny,
I am amazed at how complicated color science is and these explanations are
always an eye opener :-).
The above line of yours caught my attention and maybe you can comment.
When comparing a proof printed from an ink jet and a press print, one is
basically comparing metameric pairs, since the inks are not exactly the the
same. So each small area of the print image is a metameric pair to the
corresponding small area of the proof image. This is the first assumption.
Each of these small areas would be viewed at small angles probably much
less than 4 degrees.
Does your line suggest that there are problems in matching the print and
proof that are made from metameric pairs of small areas or have I
misunderstood your comment and taken it out of context.
Thanks.
Erik
Gerhard Fuernkranz
Graeme Gill recently pointed me to an interesting paper
Mark Q. Shaw's thesis "Evaluating the 1931 CIE Color Matching Functions"
<http://www.cis.rit.edu/mcsl/research/mshaw/CMF_Thesis.pdf>
If I understand the result of this evaluation correcly, then obviously
none of various existing CMFs are able to explain/predict a metameric
match better than with approx. 4 deltaE_a*b* average (average over a
large number of metameric sample and different persons). Isn't that a
pretty disappointing accuracy?
May I conclude that
a) trichromatic models based on a CMF weighted integration of the
spectrum simply aren't able to describe the human vision more accurately
(even with optimal CMFs)
and/or
b) the variation between various persons is probably pretty large?
I would appreciate to hear the oppinion of the experts in this group
regarding this issue.
Regards,
Gerhard
Danny Rich
"Erik Nikkanen" <nikk...@globalserve.net> wrote in message
news:d5a09$4318826e$cf7014ba$58...@PRIMUS.CA...
comparison of two small fields that are metameric can often be performed
acceptably. My post described changing the size of the image from a large
ink rollout down to the 6mm test target on the edge of the print, for
example. However, for some of the reasons cited in Mr. Fuernkranz's post,
there can and often is disagreement between human observers as to whether a
proof matches the print and more importantly on how the print differs from
the proof. This is also an example of observer metamerism.
> Thanks.
>
> Erik
>
>
>
Danny Rich
"Gerhard Fuernkranz" <nosp...@gmx.de> wrote in message
news:43188FB1...@gmx.de...
>
> Graeme Gill recently pointed me to an interesting paper
>
> Mark Q. Shaw's thesis "Evaluating the 1931 CIE Color Matching Functions"
> <http://www.cis.rit.edu/mcsl/research/mshaw/CMF_Thesis.pdf>
>
> If I understand the result of this evaluation correcly, then obviously
> none of various existing CMFs are able to explain/predict a metameric
> match better than with approx. 4 deltaE_a*b* average (average over a
> large number of metameric sample and different persons). Isn't that a
> pretty disappointing accuracy?
>
statement can be very true for some kinds of soft proofing. I gave a paper
years back - when soft proofing was just becoming of interest where a group
of individuals were asked to make matches to a series of gray metamers using
a CRT. The field was perfectly bipartite and the chromaticity and
brightness of the white points matched. All observers were tested for color
vision deficiency. In the group was an individual who looked at the
pseudo-isochromatic plates and indicated that he saw only a uniform field of
dots - no number or figures or anything. Yet his matches were undetectable
amongs the range of normal observers. But these were strongly metameric
complex gray specimens.
> May I conclude that
>
> a) trichromatic models based on a CMF weighted integration of the
> spectrum simply aren't able to describe the human vision more accurately
> (even with optimal CMFs)
>
> and/or
The 1964 observer data are much better and the modern industrial color
tolerance equations are based on those and there the agreement is very close
to the repeatability of a single observer.
>
> b) the variation between various persons is probably pretty large?
>
This is very true and often ignored.
Bob
Re:
> If I understand the result of this evaluation correctly, then obviously
> none of various existing CMFs are able to explain/predict a metameric
> match better than with approx. 4 deltaE_a*b* average (average over a
> large number of metameric sample and different persons). Isn't that a
> pretty disappointing accuracy?
Shaw and Fairchaild conclude their paper (CR&A, 2002) with this:
"...the magnitude of observer variability was nearly 8 times that of the
variability found between the responsivity functions, leading to the
conclusion that there should be more concern about the problems
introduced by observer metamerism than about the accuracy of the CIE
1931 functions themselves"
Among another troubles are the repeatedly reported breakdowns of
colorimetric additivity, which relates to your remark on the "weighted
integration of the spectrum".
In general, the publications on this converge on concluding that none of
the existing CMF is able to predict the match for any particular
observer, but they do quite well on predicting the average match of a
group of observers. This is true also for the additivity breakdowns:
fails for singles, holds in average.
b
Gerhard Fuernkranz
> Shaw and Fairchaild conclude their paper (CR&A, 2002) with this:
>
> "...the magnitude of observer variability was nearly 8 times
> that of the variability found between the responsivity functions,
> leading to the conclusion that there should be more concern
> about the problems introduced by observer metamerism than about
> the accuracy of the CIE 1931 functions themselves"
This does not sound very encouraging, particularly since most of our
color reproduction system used in practice are based on a metameric
reproduction (CRT, printer, ...).
IMO further conclusions of the results are:
I don't know the exact error probability distribution of Shaw's
evaluation, however, IMO an average error of 4dE rather suggests, that
only for a *minority* of the test persons (far below 50%) the error will
be <= 1dE.
Or in other words, even if one would select optimal metameric pairs,
which are considered as matching by as many as possible test persons
from a group, there would likely still exist an even much larger subset
of test persons in the group, which will reject these pairs as non-matching.
Or, if I can see a match between a metameric sample pair, then it is
very likely, that a *majority* of other test persons will *not* agree
with me. Thus I should not even attempt to judge a color match visually
- the result will be very likely not representative anyway for the
"average observer".
(At least for the metameric sample spectra used in Shaw's evaluation)
Danny Rich wrote:
>"Gerhard Fuernkranz" <nosp...@gmx.de> wrote in message
>news:43188FB1...@gmx.de...
>
>I would qualify that with the adjective "strongly metameric".
>
Does there exist a standardized metric to quantify terms like "slightly
metameric pair" or "strongly metameric pair"?
Particularly interesting would probably a metric, which predicts the
expected statistical color difference (e.g. expected average dE over a
large group of test persons, or even the expected error probability
distribution) for a given metameric pair of spectra, whose computed XYZ
colors are the same according to the standard observer CMFs.
(Or vice versa, a metric which solves the question, "how much" may the
spectra of a metameric pair differ, in order that a majority (or vast
majority, say 90%) of test persons would agree that the colors of the
pair match?)
Regards,
Gerhard
timo.au...@aim-dtp.net
> I'm not confident in your methodology. At 30 cm, you'd create a "patch"
> with an effective diameter of 1 cm for a 2° subtense and 5 cm for a 10°
> subtense. The two different sizes would make side-by-side comparisons
> problematic.
My understanding is that the existence of the 2deg and 10deg models in
fact says/claims that the vision sees color differently depending on
the subtense of the area. Is this not correct?
>Also, the eye sees and interprets the spectra - not an xy
> value.
What the CRT outputs is spectra, not XYZ nor Yxy nor RGB values.
>You are limited by the gamut of your monitor or your printer.
Yes, that is why I selected such spectras that result colors that are
inside the gamut of the CRT.
> The two different observation angles are meant for different
> applications. The CIE 1931 2° observer is meant for distant or small
> color sources (visual field angle subtending no more than 4°) - for
> instance, I use CIE 1931 for automotive lighting colors. The CIE 1964
> 10° observer supplement is meant for near or large color sources (visual
> field angle subtending an angle greater than 4°) - which is why it's
> used extensively in the textile and graphics industries.
I'm nor sure how to interpret the "or". In my evaluation I have small
color source side by side with a large color source, is this not per
your explanation?
> There is an interesting note in ASTM E308 for the CIE 1964 definition:
> "Users should be aware that the CIE 1964 (10°) supplementary system and
> standard observer assume no contribution or constant contribution of
> rods to vision.
But is that the same with the 1931 model also? It too is for daylight
adapted vision.
About the various metamerism issues that have been discussed in this
thread I'm not sure how valid they arey. If we simply look at the
situation:
We have two CIE models with what we can convert a spectral power
distribution be it from light source or reflected from a surface to
colorimetric CIE XYZ triplet. The spectra and it's measuring
arrangements are exactly the same. We get two different XYZ triplets
that describe two different colors that should appear the same for the
vision when the subtense of these color surfaces are accodring to the
two CIE models. Now, this surely does not happen when evaluating on the
CRT. How would metamerism related problems prevent these models to work
on the CRT? Was the 1931 and 1964 models measured using different
primaries?
Here btw are the colors from the 175 spectras side by side, both at the
same size, just to show how much different the models are
http://www.aim-dtp.net/aim/temp/2vs10.htm
The spectras were from a profiling target of a photo-exposer, the
spectral data is available in the help file of the AIM.XLA
http://www.aim-dtp.net/aim/technology/aim_xla/index.htm
AIM.XLA was used for the calculations.
Timo Autiokari
bob
> Particularly interesting would probably a metric, which predicts the
> expected statistical color difference (e.g. expected average dE over a
> large group of test persons, or even the expected error probability
> distribution) for a given metameric pair of spectra, whose computed XYZ
> colors are the same according to the standard observer CMFs.
>
> (Or vice versa, a metric which solves the question, "how much" may the
> spectra of a metameric pair differ, in order that a majority (or vast
> majority, say 90%) of test persons would agree that the colors of the
> pair match?)
Such metric exists - the Standard Deviate Observer (CIE SDO). However,
it is not really doing its job and can possibly only be used to compare
the levels of metamerism between two pairs; the absolute SDO values
have no much meaning.
b
Danny Rich
<timo.au...@aim-dtp.net> wrote in message
news:1125899214....@g44g2000cwa.googlegroups.com...
Douglas G. Cummins wrote:
> I'm not confident in your methodology. At 30 cm, you'd create a "patch"
> with an effective diameter of 1 cm for a 2° subtense and 5 cm for a 10°
> subtense. The two different sizes would make side-by-side comparisons
> problematic.
My understanding is that the existence of the 2deg and 10deg models in
fact says/claims that the vision sees color differently depending on
the subtense of the area. Is this not correct?
***
The answer to this is YES and NO. White is white and the white points are
very close to each other. But other colors - especially in the blues - will
be perceived quite differently.
***
>Also, the eye sees and interprets the spectra - not an xy
> value.
What the CRT outputs is spectra, not XYZ nor Yxy nor RGB values.
*** But what you see are trichromatic sensations ***
>You are limited by the gamut of your monitor or your printer.
Yes, that is why I selected such spectras that result colors that are
inside the gamut of the CRT.
> The two different observation angles are meant for different
> applications. The CIE 1931 2° observer is meant for distant or small
> color sources (visual field angle subtending no more than 4°) - for
> instance, I use CIE 1931 for automotive lighting colors. The CIE 1964
> 10° observer supplement is meant for near or large color sources (visual
> field angle subtending an angle greater than 4°) - which is why it's
> used extensively in the textile and graphics industries.
I'm nor sure how to interpret the "or". In my evaluation I have small
color source side by side with a large color source, is this not per
your explanation?
> There is an interesting note in ASTM E308 for the CIE 1964 definition:
> "Users should be aware that the CIE 1964 (10°) supplementary system and
> standard observer assume no contribution or constant contribution of
> rods to vision.
But is that the same with the 1931 model also? It too is for daylight
adapted vision.
*** Again there is a YES and NO here. The observations were not carried out
at day time illumination levels.
About the various metamerism issues that have been discussed in this
thread I'm not sure how valid they arey. If we simply look at the
situation:
We have two CIE models with what we can convert a spectral power
distribution be it from light source or reflected from a surface to
colorimetric CIE XYZ triplet. The spectra and it's measuring
arrangements are exactly the same. We get two different XYZ triplets
that describe two different colors that should appear the same for the
vision when the subtense of these color surfaces are accodring to the
two CIE models. Now, this surely does not happen when evaluating on the
CRT. How would metamerism related problems prevent these models to work
on the CRT? Was the 1931 and 1964 models measured using different
primaries?
***
Obtaining two different XYZ triplets for the same spectral stimuli is the
inverse definition of metamerism. And yes - the 1931 and 1964 experiments
used different sets of primaries.
***