Out of the Wood

BY MIKE WOOD

CRI and the Color Quality Scale, Part 2

CQS offers discerning users a metric to allow direct comparison
of luminaires with different light source technology

In the Winter 2010 issue of Protocol I wrote about the source would render saturated reds and purples very poorly while
Color Rendering Index (CRI) and how it is calculated. That over emphasizing saturated blues. (The blue line in Figure 2 is
article finished with a brief discussion on how the CRI value the reference while the red is the calculated result under the test
and measurement is perhaps a poor one for assessing arrays of light source. If the light source were perfect the red and blue lines
colored LEDs. would coincide.)

The problems with CRI

To recap, the normal CRI value is based on the calculated ability of
a light source to render eight standard colors. The eight colors are
all relatively unsaturated, which works well for broad band light
sources with continuous spectra, but can be problematic for narrow
band LED sources with large peaks and valleys in their spectra. An
RGB light source can produce good rendering of the unsaturated Figure 1 -
test colors, resulting in a high CRI, even when its rendering of Spectra of
saturated colors is poor. Another problem is that the CRI is RGB LED with
CRI of 81
calculated as a simple average of the rendering of the eight colors.
This makes it possible for a light source to obtain a high CRI even
though it renders one or two of the colors very poorly. This is often
the case with RGB LEDs where the precise selection of wavelengths Now let’s just move those
chosen for the three colors and how they match up with the eight LED wavelengths very slightly
test colors can be critical. A change of a few nanometers in color to 455 nm, 534 nm and 616
of an emitter can swing the CRI from 70 to 90. This is a purely nm (RGB 2) as shown in
artificial swing related to inadequacies in the measuring technique Figures 3 and 4. The result
and the results are misleading. LED RGB triads may look similar to of this small change is a huge
the eye but give wildly varying CRI results. This problem also opens drop in CRI down to 67, which
the possibility of a manufacturer gaming the CRI of a product by is a level most people would
carefully picking LED wavelengths that result in a high CRI. say was unacceptable. However
Let’s look at an example of this. Figures 1 and 2 show the spectra Figure 2 - Color Rendering of RGB a careful look at Figure 4
and color rendering of an RGB LED source, let’s call it RGB 1, LED with CRI of 81 reveals that, in fact, most colors
with LED wavelengths centered at 460 nm, 540 nm and 605 nm. are better rendered than in the sample with a CRI of 81, the biggest
At a nominal CCT of 3300 K this source has a calculated CRI of errors are in the green and red where colors are over emphasized,
81 which is quite high and normally would be considered very and that nowhere in the gamut are any colors under-rendered. Most
respectable. However, if you look at Figure 2 you can see that this people would prefer this light source even though it has a low CRI.
S P RI N G 2 01 0

14
S P R I NG 2 0 1 0
 Out of the Wood | CRI and the Color Quality Scale, Part 2

The CQS, like the CRI, is a test sample method. That is, color
differences are calculated for a standard set of colored samples
when illuminated by the test source and a reference illuminant.
As mentioned above the CRI samples are all relatively unsaturated
colors and this can hide problems a source may have rendering
more saturated tones. NIST has established through extensive
computational testing that, although light sources can perform
poorly with saturated samples even when performing well with
Figure 3 - Spectra
unsaturated ones, the inverse is never true. That is, there is no light
of RGB LED with
CRI of 67 source spectrum that would
render saturated colors well,
and render unsaturated
Note: This overemphasizing of colors poorly. This important
some colors is common with result shows that nothing is
narrow band light sources like lost and everything is gained
LEDs and can lend a cartoon by only using saturated
like or hyper-real appearance colors as our new sample set.
to colored objects. Personally I Therefore, CQS uses fifteen
don’t like this, as even though it saturated colors chosen to
is less overtly objectionable than be evenly spaced across the
under-rendering, it is actually just entire visible spectrum.
as much of a problem as under
rendering for entertainment
lighting where color fidelity is
Figure 4 - Color Rendering of
RGB LED with CRI of 67 often the goal. Figure 5 - CQS Standard Colors

Color Quality Scale

In recognition of these problems NIST (the National Institute of
Standards and Technology) has been working on a new means for
measuring and reporting color rendition called the Color Quality
Scale (CQS). The goal was to keep the good points of CRI with
its use of standard color chips and direct relation to the real-
world, while addressing the shortcomings arising from the choice
of standard colors and the math used to combine the results.
A major decision in the new metric was to continue to report
results as a single number. Although this inevitably results in some
compromises in the resolution of the results, it was felt important
to keep that link to the well known and understood CRI. The
purpose of a metric like CQS is to condense an immense amount
of information into something manageable and useful. In order to
be useful for the greatest number of users, most of whom have very
limited knowledge of colorimetry, a one-number output continues
to be desirable. Throughout our personal and professional lives
we use many measurement scales whose precise meanings and
measurement methodologies are unknown to us without concern.
Examples of such scales include shoe sizes, octane ratings of gasoline,
and radio station frequencies. Though most people don’t know
precisely how those numbers are determined, they find the scales
useful and have a general understanding of how different outputs
relate to each other (a larger shoe size means a bigger foot!).
S PR I NG 2 0 10

15
pr oto c o l
 Out of the Wood | CRI and the Color Quality Scale, Part 2

Compare Figure 5 with the sample set used for CRI (Figure 1 in for CRI, there are a number of important differences between how
the Winter 2010 issue of Protocol) and you can see how much more those values are used to calculate the final metric.
saturated these are than the TCS01 - TCS08 samples typically used I wrote earlier that the simple averaging of the color difference
for CRI. Figure 6 shows the full set of CQS test colors and their values, as happens with CRI, can result in assigning a source a high
spectra. (Note: These colors are unlikely to appear accurately in this CRI value even though one or two samples show significant color
journal. The limitations of the printing process will render them as less differences. The CQS avoids this by combining the 15 values by
saturated and with different tonal values than the originals.) an RMS (root-mean-square) calculation. By squaring every value
All fifteen CQS colors are available as real samples with standard before averaging them we emphasize any errors and ensure that
Munsell numbers but, as with CRI, there is no need to ever use poor rendering of even a few of the samples will have a significant
them! Everything you need to calculate CQS can be derived from impact on the result. There are other changes in the math for CQS
the source spectrum and knowledge of the color properties of that further improve the result over that of CRI, but these are out of
the samples. Although the initial calculation of the errors in the the scope of this article. However the result, I believe, is something
rendering of each of the fifteen colors is very similar to that used that will suit the entertainment business very well and will give us a
true metric for how good a light source’s color rendering is, both to
Figure 6 - CQS Test Color Spectra the human eye and to the TV or film camera.

0.6 0.6 0.6

CQS Test Color Sample 1 CQS Test Color Sample 2 CQS Test Color Sample 3
0.5 0.5 0.5

0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
350 450 550 650 750 300 400 500 600 700 800 300 400 500 600 700 800
Wavelength, nm Wavelength, nm Wavelength, nm

0.6 0.6 0.6

CQS Test Color Sample 4 CQS Test Color Sample 5 CQS Test Color Sample 6
0.5 0.5 0.5

0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
300 400 500 600 700 800 300 400 500 600 700 800 300 400 500 600 700 800
Wavelength, nm Wavelength, nm Wavelength, nm

0.8 0.8 0.8

0.7 CQS Test Color Sample 7 0.7 0.7 CQS Test

T t Color
C l Sample
S l 9
CQS Test Color Sample 8
0.6 0.6 0.6

0.5 0.5 0.5

0.4 0.4 0.4

0.3 0.3 0.3

0.2 0.2 0.2

0.1 0.1 0.1

0 0 0
300 400 500 600 700 800 300 400 500 600 700 800 300 400 500 600 700 800
Wavelength, nm Wavelength, nm Wavelength, nm

0.8 0.8 0.8

0.7 CQS Test Color Sample 10 0.7 CQS Test Color Sample 11 0.7 CQS Test Color Sample 12
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
0.1 0.1 0.1
0 0 0
300 400 500 600 700 800 300 400 500 600 700 800 300 400 500 600 700 800
Wavelength, nm Wavelength, nm Wavelength, nm

0.8 0.8 0.8

0.7 CQS Test Color Sample 13 0.7 CQS Test Color Sample 14 0.7 CQS Test Color Sample 15
0.6 0.6 0.6
0.5 0.5 0.5
0.4 0.4 0.4
0.3 0.3 0.3
0.2 0.2 0.2
S P RI N G 2 01 0

0.1 0.1 0.1

0 0 0
300 400 500 600 700 800 300 400 500 600 700 800 300 400 500 600 700 800
Wavelength, nm Wavelength, nm Wavelength, nm

16
S P R I NG 2 0 1 0
 Out of the Wood | CRI and the Color Quality Scale, Part 2

Example results areas, and a second triad, RGB 2, which had a poor CRI of 67, but
actually did a better job in many areas. Running them through the
Let’s take a look at the CQS results for some real light sources to see
CQS calculations we get results of 75 for RGB 1 and 79 for RGB 2.
how they stack up. Figure 7 shows an incandescent lamp.
Figure 9 shows the CQS samples when illuminated by RGB 2 where
you can see the over-emphasis or chroma-enhancement of the red,
Figure 7 - CQS samples under an incandescent lamp amber, and green.

Figure 9 - CQS samples under RGB LEDs at 455nm, 534nm and 616nm

Perhaps surprisingly an incandescent doesn’t have a CQS of 100.

Instead, it is 98. A real incandescent lamp is not a perfect black
body emitter as incandescent lamps are usually slightly inadequate The CQS values for RGB 1 and RGB 2 are now much closer
in the blue, but CQS has the ability to recognize and report that. together than the CRI, as one would intuitively expect, and CQS
(The color differences are too small to be rendered in this image— correctly penalizes RGB 1 for poor color rendering in a small area.
it all looks perfect here!)
Figure 8 shows a mercury lamp. There are errors throughout
the whole spectrum with the largest in the blues and yellows. As
expected this lamp has a poor CQS of 46.

Figure 8 - CQS samples under a mercury lamp

But what about those two hypothetical LEDs we talked about

earlier? How do they compare when we use the CQS metric instead
of CRI? If you recall we had one LED triad, RGB 1, which had a
CRI of 81, even though it had very poor color rendering in some
S PR I NG 2 0 10

17
pr oto c o l
Out of the Wood | CRI—What does it really mean?

This is a much more palatable and representative result. However, strictly. In the case of our hypothetical RGB 1 and RBB 2 LEDs
it’s not perfect. The standard CQS calculation recognizes that this results in an unchanged Color Fidelity result of 75 for RGB 1
over-emphasizing a color is often less objectionable than under- whereas RGB 2 (the over-emphasizer) drops down to 71. Both these
rendering, so it penalizes errors from over-rendering less severely. values seem to better realistically represent what the eye sees with
Sometimes that’s true in our industry too, but I suspect that often narrow-band emitters than does CRI and give us a much better idea
over-emphasis isn’t acceptable, as it always causes associated errors of what to expect when comparing these narrow-band sources with
in hue. Thus we want to penalize over-emphasis in the metric. traditional, broad-band sources.
Fortunately, CQS offers a solution. NIST is still working on testing and developing CQS but I
Although we mentioned earlier that CQS is a one-number believe it’s a metric we should look at adopting for entertainment
metric, NIST acknowledges that certain applications require more lighting luminaires. We know CRI does a poor job, and with LEDs
specific information about the color rendering properties of light is inadequate and often misleading. CQS however should give us
sources, and I would argue that entertainment lighting is one of a metric that will allow users to directly compare luminaires with
those applications. We use color extensively in very creative and different light sources and get results that make sense no matter
precise ways and color accuracy is of profound importance to many what the light source technology. n
designers. CQS offers discerning users additional indices, one of
which I think is particularly relevant to our industry. Credits: Many thanks to Wendy Davis at NIST for permission to
reproduce text and figures from NIST documents.

Color Fidelity Scale Mike Wood is President of Mike Wood Consulting LLC which p r o v i d e s
This extra metric is the Color Fidelity Scale. It is intended, as its c onsulting support to c ompanies within the entertainment indu stry on
tec hnology strategy, R& D, standards, and I ntellec tual Property. A 30- ye a r
name suggests, to evaluate the fidelity of object color appearances.
veteran of the entertainment technology industry, Mike is the Tr e a s u r e r a n d
It removes the leniency accorded to over-emphasis of colors from I mmediate Past President of ESTA. Mike c an be reac hed at 512. 288. 4916.
the main CQS calculation and reports errors of any kind equally