Colour Matching Experiments

with RGB-LEDs

P. Csuti, J. Schanda*
Virtual Environments and Imaging Technologies Laboratory, University of Pannonia, Veszprém, Hungary

Received 17 March 2007; revised 30 April 2007; accepted 1 June 2007

Abstract: CIE colorimetry breaks down when lights pro- In a second experiment, red, green, and blue light
duced by narrow band RGB-LEDs are matched with broad- emitting diodes (RGB-LEDs) were used as primaries, and
band lights. A colour matching experiment was set up and Maxwell matches were made between mixtures of the RGB-
matches in a number of parts of the chromaticity diagram LED lights and broadband stimuli.6 In that case, differences
have been made, to determine the magnitude of the discrep- were found, and we could show that if the colour matching
ancy. Differences between visual and instrumental matches functions suggested by Vos7 were used in place of the CIE
increase as one moves in the chromaticity diagram from 1931 colour matching functions (CMFs), better agreement
yellowish white lights toward greenish and bluish lights. between the visual and instrumental data could be obtained.
CIE TC 1-36 recently suggested newly defined cone Work started a few years ago in the CIE (CIE TC 1-36)
fundamentals: Applying a transformation of these to a to obtain a physiologically based chromaticity diagram.8
space similar to the CIE XYZ space enables a much better This work was partially driven by the 50-year-old knowl-
prediction of the matches to be made. The difference edge that the CIE 1931 CMFs are in error9 because the
between the visual match and its instrumental prediction incorporated CIE 1924 spectral luminous efficiency function
decreases by a factor of two or even more. [V(k)] runs too low in the blue part of the spectrum. An
The use of a cone fundamental based colorimetric system other CIE Technical Committee has begun investigations
is recommended for LED colorimetry. Ó 2008 Wiley Periodi- of the validity limits of Grassmann’s laws, and in particular
cals, Inc. Col Res Appl, 33, 108 – 112, 2008; Published online in Wiley
of the question of error propagation in the transformations
InterScience (www.interscience.wiley.com). DOI 10.1002/col.20385
of the primaries.10,11 At the suggestion of this committee,
several laboratories started colour matching experiments
Key words: colour matching; cone fundamentals; RGB-LED and reported deviations from the CIE CMFs.12–14
The L, M, S cone fundamentals published by the CIE
TC 1-368 do not transform to the CIE 1931 CMFs. As
these cone fundamentals reflect present day knowledge of
INTRODUCTION average human colour vision,15 it seemed to be reasonable
CIE colorimetry was conceived to predict colour matches to test these, to find how well they describe visual meta-
of stimuli with different spectral distributions. Over the meric matches. Wold16 has prepared an LMS–XYZ trans-
years, several experiments have been described that pointed formation. Based on these calculations CIE TC 1-36
published equations into a draft technical report for trans-
toward a break down of metameric colour matches (see an
forming LMS cone fundamentals in CIE XYZ like CMFs.17
early summary e.g. Ref. 1). Thornton reported the most dra-
Equation 1 shows the matrix transformation equation to
matic differences between observed and calculated colour
transform the CIE TC 1-36 LMS cone fundamentals8 into
matches in his three-part paper.2–4 In an earlier paper, one XYZ like CMFs.*
of the present authors reported on metameric matches      
     
between broad-band stimuli and stimuli produced on a cath-  xF ðlÞ   1:910988 1:394658 0:389317   lðlÞ 
       (1)
ode ray tube monitor,5 because this is now of importance  yF ðlÞ  ¼  0:643151 0:395946 0:000000 
 mðlÞ 
when many visual experiments are done on computer-con- zF ðlÞ   0:000000 0:000000 1:919339    sðlÞ 
trolled monitors. Experiments showed that for such stimuli
the CIE colorimetry holds well.
*The tabulated values of these functions can be found at http://
vision.vein.hu/schanda/Manuscripts/CMFs. The authors stress that the
*Correspondence to: J. Schanda (e-mail: [email protected]).
reproduced CMFs are not CIE endorsed CMFs, but show an interim result
V
C 2008 Wiley Periodicals, Inc. of the TC.

108 COLOR research and application

 TABLE I. Luminance and chromaticity of the nine
test locations.
Chromaticity
Luminance coordinates
Test location (cd/m2) x y

1. CC2-BG2 169.9 0.322 0.327

2. C3C20-BGG20 88.2 0.158 0.259
3. C3C22-BGG22 182.7 0.174 0.378
4. 3C3-GG3 129.4 0.248 0.514
5. 3C11-GG11 151.8 0.294 0.588
6. :3C9-YGG9 401.7 0.428 0.512
7. PC7-PG7 394.3 0.443 0.366
8. PC8-PG8 106.4 0.656 0.325
9. TC10-RG10 108.1 0.525 0.425

fuser via a greenish-blue filter, and in the left compart-

ment the mixture of three LEDs illuminate the diffuser
FIG. 1. Front view of the experimental box. (colour match has not yet been established). The box was
set up in a dark room, with a white diffuser behind it.
Thus, it became possible to compare the performance This was illuminated by an incandescent lamp producing
of the CIE 28 observer with a 28 observer built on the an adaptation luminance of about 160 cd/m2. No spill
most up-to-date cone fundamental data. Pilot study Max- light reached from this illuminated diffuser, the test and
well match results were published at the CIE 2006 Expert reference windows of the test box.
Symposium in Ottawa,18 and comparative results obtained High-power red, green, and blue LEDs were placed
in Veszprém and in Ilmenau at the Paris CIE Symposium into the test compartment, with dominant wavelengths of
in 2006.19 The present article provides a more compre- 626 nm, 525 nm, and 476 nm, respectively. (The domi-
hensive review of the results obtained both for achromatic nant wavelengths of the red and blue LEDs were near to
and chromatic colours. Thornton’s antiprime wavelength, the green, however,
was closer to the prime colour wavelength.) Figure 2
shows the relative spectral power distribution of the LEDs
EXPERIMENTAL EQUIPMENT used in this experiment. A PC powered the LEDs via a
A double compartment illuminator was built, where on microcontroller system that enabled the colour to be
one side (reference field) an incandescent lamp illumi- changed according to its three perceptual attributes: hue,
nated a holographic diffuser, which was seen as a 2 3 3 brightness, and saturation. This made the setting of the
degree visual field with an observation distance of 1 m. match much easier, compared with a system where the
Optical glass filters could be put in front of the light intensities of the three RGB-LEDs had to be changed.
source to modify the colour of the light of this reference Russian 40 mm 3 40 mm-sized colour glasses were
field. The luminance of the reference field depended on usedy to change the spectral power distribution of the in-
the colour of the filters, but was in the range of 90–400 candescent lamp. After an observer has signaled colour
cd/m2 for the different colours used. The test field of sim- match, the colour and luminance of the test and reference
ilar size was placed beside of the reference field but was windows were measured using a Photo Research PR 705
separated from it by a black stripe 10 arc minutes wide, spectroradiometer. Calibration of the instrument was per-
as can be seen in Fig. 1. In the example shown, in the formed at the University of Ilmenau using a PTB cali-
right compartment the halogen lamp illuminates the dif- brated FEL lamp.{ Based on these comparative measure-
ments, and after correction of the systematic errors of the
instrument, a maximum uncertainty of one to two digits
in the third decimal of the chromaticity coordinates could
be obtained. Nine locations in the chromaticity diagram
have been selected for the present test. Table I shows the
luminance and measured 28 chromaticity values of the
reference field.

y
The transmission characteristics of the colour filters can be found at
http://vision.vein.hu/schanda/Manuscripts/CMFs. In the paper, the
authors provide the original Cyrillic designation of the glasses together
with an English translation: BG, blue glass; BGG, blue-green glass; GG,
green glass; YGG, yellowish green glass; PG, pink glass; RG, red glass.
{
FIG. 2. Relative spectral power distributions of the LEDs The authors thank Ms. Bieske of the University of Ilmenau and Dr.
tested. Krüger of TechnoTeam for their help in the calibration.

Volume 33, Number 2, April 2008 109

TABLE II. Inter- and intra-observer standard
deviation of the chromaticities.
Standard deviation of the test
chromaticities (u0 , v0 )
Inter-observer Intra-observer
Test location (n ¼ 6) (k ¼ 5)

1. CC2-BG2 0.0029 0.0005

2. C3C20-BGG20 0.0016 0.0009
3. C3C22-BGG22 0.0048 0.0006
4. 3C3-GG3 0.0034 0.0010
5. 3C11-GG11 0.0018 0.0009
6. :3C9-YGG9 0.0043 0.0014
7. PC7-PG7 0.0025 0.0012
8. PC8-PG8 0.0042 0.0024
9. TC10-RG10 0.0034 0.0017

EXPERIMENTS AND RESULTS

Six young colour normal observers (five males and one
female) took part in the visual experiments. Every ob-
server performed five matches for one experimental set-
up, and the average of these has been recorded, together
with their standard deviations.
The repeatability of the visual settings was comparable or
slightly larger than the uncertainty of the instrumental mea-
surement. Table II shows the inter- and intra-observer stand-
ard deviations, calculated in u0 , v0 coordinates so that better
comparisons could be made between different locations in FIG. 3. Measured values of the chromaticities of the test
the chromaticity diagram. The uncertainties of one observer and reference fields in case of colour match for the nine
test sites and the average of the six observers in the u0 , v0
in setting the colour match (intra-observer uncertainty) are diagram, using the CIE 28 standard observer (a), and the
roughly half as big as the inter-observer uncertainty. CMFs derived from the visual fundamentals (b).
The results of the colour matches are shown in Fig. 3.
The upper figure shows the results obtained if the spectra Wold (personal communication) were used instead of
were calculated using the CIE 1931 standard observer. those of the CIE 1931 standard observer.
Open circles show the average values of the results of the
six observers, and filled circles show the average of all
visual data. The filled in triangle shows the chromaticity DISCUSSION
measured on the filtered incandescent lamp reference Comparing the upper and lower parts of Fig. 3, it can be
field. In the lower diagram, the chromaticities were plot- seen that the measured chromaticity differences between
ted with if the CMFs derived from the fundamentals by the RGB-LED and the filtered incandescent lights are

FIG. 4. Enlarged view of the u0 , v0 chromaticity diagram in the vicinity of sample #1; (a) CIE 28 standard observer; (b) fun-
damental based 28 CMFs.

110 COLOR research and application

FIG. 5. Enlarged view of the u0 , v0 chromaticity diagram in the vicinity of sample #5; (a) CIE 28 standard observer, (b) fun-
damental based 28 CMFs.

smaller in the case of the CMFs derived from the cone By carrying out an ANOVA test to investigate the
fundamentals. As can be seen the CIE CMFs describe the overall influence of the three types of fundamentals used
metameric match quite well in the red to yellow part of in the difference calculations (Table III) a significant
the chromaticity diagram, but the differences become effect was found [F(2, 159) ¼ 6.325, P ¼ 0.002]. Further
larger in the green and blue regions. analysis using the least significant difference post hoc test
Figure 4 shows the enlarged view of the u0 , v0 chroma- showed that there was a significant difference separately
ticity diagram in the vicinity of sample #1, for the CIE between using the CIE CMFs and the LMS derived CMFs
CMFs in Fig. 4(a), and for the diagram derived from the and between the CIE CMFs and the Vos fundamental ob-
cone fundamentals in Fig. 4(b). Figure 5 shows similar server, but no significance was found for the comparison
pictures for sample #2. of the Vos fundamental observer and LMS-derived CMFs.
The chromaticity differences for all nine samples are to As the cone fundamental based CMFs were derived from
be seen in Table III. Here the results obtained with the physiological data that take more recent functions with
CIE 1931 CMFs and the CMFs derived from the cone high reliability into consideration—and although nonsigni-
fundamentals are shown, together with those from an ear- ficantly, but in almost every case (exception #6) the cone
lier experiment involving the Vos fundamental observer. fundamental based CMFs provided the better agreement
As in the earlier experiment these provide better results with the visual observations than the Vos fundamental ob-
than the CIE CMFs.6 The standard deviation of the nine server—one can recommend the cone fundamental
chromaticity differences is also shown. For every test derived CMFs for further testing and future use.
location, the lowest value is shown in bold italic. Figure
6 shows the differences in a graphical form. CONCLUSIONS

Highly metameric colour matches were performed

TABLE III. CIE 1976 chromaticity difference between
between colours provided by mixing the light of RGB-
the filtered incandescent light and the metameric
RGB cluster for the three CMFs: CIE 28 observer,
Vos fundamental observer, and cone fundamental
derived CMFs.
CIE 28 Vos Cone
standard fundamental fundamental
Test location observer observer derived CMFs

1. CC2-BG2 0.0253 0.0118 0.0113

2. C3C20-BGG20 0.0383 0.0184 0.0129
3. C3C22-BGG22 0.0249 0.0110 0.0104
4. 3C3-GG3 0.0126 0.0062 0.0053
5. 3C11-GG11 0.0026 0.0023 0.0022
6. :3C9-YGG9 0.0017 0.0027 0.0031
7. PC7-PG7 0.0173 0.0094 0.0093
8. PC8-PG8 0.0024 0.0067 0.0026
9. TC10-RG10 0.0062 0.0056 0.0040 FIG. 6. CIE 1976 chromaticity difference between the fil-
Standard deviation 0.0128 0.0050 0.0042 tered incandescent light and the metameric RGB cluster
for the three CMFs: CIE 28 observer, Vos fundamental ob-
Lowest value for each test location is bold italic. server and cone fundamental derived CMFs.

Volume 33, Number 2, April 2008 111

LEDs (dominant wavelengths: 626 nm, 525 nm, and 476 7. Vos JJ. Colorimetric and photometric properties of a 28 fundamental
nm) and filtered incandescent lights. Large visual mis- observer. Color Res Appl 1978;3:125–128.
8. CIE Technical Report. Fundamental chromaticity diagram with phys-
matches were observed—especially in the blue part of the iological axes. I. Publ CIE 2006. p 170–171.
chromaticity diagram—if instrumental colour match was 9. Judd DB, Secretariat report: Colorimetry and artificial daylight. In
set between the two lights. Proceedings of the 12th Session CIE, Stockholm 1, Techn. Com. No.
On making visual matches between the filtered incan- 7. Publ. CIE Central Bureau, 1951.
descent and the RGB-LED lights, and measuring the spec- 10. CIE TC 1–56. Com: Improved Colour Matching Functions.
11. Brill MH. Towards resolving open questions on the validity of
tral power distribution of the two lights, the calculated Grassmann’s laws. In Proceedings of ISCC/CIE Expert Symposium
chromaticities differed considerably. Differences in the ’06—75 years of the CIE standard colorimetric observer. Publ. CIE
second decimal of the u0 , v0 chromaticities were found. If x030, 2006. p 8–12.
instead of the CIE 28 colour matching functions derived 12. Oleari C, Pavesi M. New measures of the color-matching functions
from cone fundamentals were used, the calculated chro- in foveal vision. In Proceedings of the 10th Congress AIC, AIC Col-
our 05, Granada, 2005. p 599–602.
maticity differences became smaller by a factor of 2 to 3. 13. Oicherman B, Luo MR, Robertson AR, Tarrant A. Experimental ver-
As lights produced by mixing the light of red, green, ification of tri-chromatic generalisation. In Proceedings of 10th Con-
and blue LEDs becomes general practice, the use of the gress AIC, AIC Colour 05, Granada, 2005 p 29–32.
CMFs derived from cone fundamentals instead of the CIE 14. Oicherman B, Luo MR, Robertson AR. Test of the transformation of
CMFs is to be recommended to avoid visual chromaticity tristimulus space: Forward- and inverse-matrix methods. In Proceed-
ings of ISCC/CIE Expert Symposium ’06—75 years of the CIE
differences. standard colorimetric observer. Publ. CIE x030, 2006. p 30–36.
15. Stockckman A, Sharpe L. Physiologically-based colour matching func-
tions. In Proceedings of ISCC/CIE Expert Symposium ’06—75 years of
1. Robertson AR. Overview of sixty years of CIE colorimetry. Publ. the CIE standard colorimetric observer. Publ. CIE x030, 2006. p 13–20.
X007. Proceedings of CIE Symposium on Advanced Colorimetry. 16. Wold JH. XYZ representations of the Stockman-Sharp-Fach cone
Vienna, Austria, 1993. p 3–6. fundamentals. Color Res Appl 2008 (submitted for publication).
2. Thornton WA. Towards a more accurate and extensible colorimetry. 17. CIE TC 1-36 draft technical report of Chapter 7.3: Development of
Part I. Introduction, the visual colorimeter-spectroradiometer. Experi- chromaticity diagrams based upon the principles of the CIE XYZ
mental results. Color Res Appl 1992;17:79–122. system (draft 2006).
3. Thornton WA. Towards a more accurate and extensible colorimetry. 18. Csuti P, Schanda J. Colour matching based on fundamental spectral
Part II. Discussion. Color Res Appl 1992;17:162–186. sensitivity functions. In Proceedings of ISCC/CIE Expert Symposium
4. Thornton WA. Towards a more accurate and extensible colorimetry. ’06—75 years of the CIE standard colorimetric observer. Publ. CIE
Part III. Discussion continued. Color Res Appl 1992;17:240–262. x030, 2006. p 21–24.
5. Ábrahám GY, Körösi H, Schanda J, Shapiro AG, Wenzel K. Anoma- 19. Bieske K, Csuti P, Schanda J. Colour appearance of metameric white
lies in additive colour matches. Color Res Appl 1995;20:235–244. lights and possible colorimetric description. In Publ. CIE x032 Pro-
6. Borbély Á, Schanda J. Colour matching using LEDs as primaries. ceedings of the 2nd CIE Expert Symposium–Visual Appearance,
Color Res Appl 2004;29:360–364. 2007. p 137–139.

112 COLOR research and application