UNIVERSITY OF JOENSUU

DEPARTMENT OF PHYSICS
VÄISÄLÄ LABORATORY

DISSERTATIONS 30

Accurate Color Measurement

Jouni Hiltunen

ACADEMIC DISSERTATION

To be presented, with permission of the Faculty of Science of the Univer-

sity of Joensuu, for public criticism in Auditorium M1 of the University,
Yliopistonkatu 7, Joensuu, on February 8th, 2002, at 12 noon.

JOENSUU 2002
Julkaisija Joensuun yliopisto
Publisher University of Joensuu

Toimittaja Timo Jääskeläinen, Ph.D., Professor

Editor

Ohjaajat Timo Jääskeläinen, Ph.D., Professor

Supervisors Department of Physics, University of Joensuu
Jussi Parkkinen, Ph.D., Professor
Department of Computer Science, University of Joensuu
Joensuu, Finland

Esitarkastajat Mauri Aikio, Dr. Tech.

Reviewers VTT Electronics
Oulu, Finland
Erik Vartiainen, Ph.D., Docent
Department of Electrical Engineering,
Lappeenranta University of Technology
Lappeenranta, Finland

Vastaväittäjä Harri Kopola, Dr. Tech., Professor

Opponent VTT Electronics
Oulu, Finland

Vaihto Joensuun yliopiston kirjasto, vaihdot

PL 107, 80101 JOENSUU
Puh. 013-251 2677, telefax 013-251 2691
Email: vaihdot@joensuu.fi
Exchange Joensuu University Library, exchanges
P.O. Box 107, FIN-80101 JOENSUU
Telefax +358 13 251 2691
Email: vaihdot@joensuu.fi

Myynti Joensuun yliopiston kirjasto, julkaisujen myynti

PL 107, 80101 JOENSUU
Puh. 013-251 2652, 251 2677, telefax 013-251 2691
Email: joepub@joensuu.fi
Sale Joensuu University Library, sale of publications
P.O. Box 107, FIN-80101 JOENSUU
Telefax +358 13 251 2691
Email: joepub@joensuu.fi

ISSN 1458-5332
ISBN 952-458-077-2

Joensuun yliopistopaino 2002

Jouni Hiltunen∗ ; Accurate Color Measurement – University of Joensuu, Depart-
ment of Physics, Väisälä Laboratory, Dissertation 30, 2002. - 88 p.
ISBN 952-458-077-2
Keywords: accurate color measurement, tristimulus integration, thermochromism.
∗
Address: Department of Physics, University of Joensuu, P.O. Box 111, FIN-80101,
Joensuu, Finland

Abstract
In this thesis studies on accurate surface color measurements are considered. Errors in
a spectrophotometric measurements are discussed and correction methods introduced.
The wavelength interval in a tristimulus integration is considered next. The ASTM
weighting functions are tested with a large data set of non fluorescent colors and shown
to be useless. The thermochromic effect is also discussed in detail and measured. This
thesis shows how thermochromism is based on physical processes. Simple formulas are
derived, and shown to explain the experimental data. In conclusion, this thesis shows
how commercial instruments should be calibrated for precision color measurements, if
one aims at achieving the highest level for measuring accuracy.
iv

Preface
I am deeply indebted to my supervisors Prof. Timo Jääskeläinen and Prof. Jussi
Parkkinen for their guidance, encouragement and patience during my studies. I am
grateful for the opportunity to work at the Department of Physics.
I want to express my special thanks to all my co-workers during these years, espe-
cially Merja, Kimmo and Jarkko. Furthermore, I wish to express my gratitude to the
colleagues and staff of the department, and the members of the color group.
To my referees, Dr. Mauri Aikio and Docent Erik Vartiainen, I am greatly indepted
for their careful review and constructive comments. For revising the language of the
manuscript I express my gratitude to Dr. Greg Watson.
Finally, I want to express my warmest thanks to my dear parents and to my sister
with her family, and to Liisa for her love and support.

Joensuu January 23, 2002

Jouni Hiltunen
 Contents

1 Introduction 1

2 Colorimetry 4
2.1 Background to colorimetry . . . . . . . . . . . . . . . . . . . . . . . . . 4
2.2 CIE Recommendations . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.1 Standard physical data . . . . . . . . . . . . . . . . . . . . . . . 7
2.2.2 Standard observer data . . . . . . . . . . . . . . . . . . . . . . . 12
2.2.3 Calculation of tristimulus values and chromaticity coordinates . 12
2.2.4 Uniform color spacing . . . . . . . . . . . . . . . . . . . . . . . 15
2.2.5 Miscellaneous colorimetric practices and formulae . . . . . . . . 17
2.3 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

3 Measurement of surface color 21

3.1 General considerations on color measurement . . . . . . . . . . . . . . . 21
3.2 Intercomparison . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.3 Errors in surface color measurements . . . . . . . . . . . . . . . . . . . 24
3.3.1 Errors in absolute scales of diffuse reflectance and 0/45 radiance
factor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.3.2 Errors due to differing properties of white reference standards . 25
3.3.3 Photometric non-linearity . . . . . . . . . . . . . . . . . . . . . 25
3.3.4 Incorrect zero level . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.5 Wavelength scale error . . . . . . . . . . . . . . . . . . . . . . . 27
3.3.6 Specular beam exclusion error . . . . . . . . . . . . . . . . . . . 28
3.3.7 Specular beam weighting error . . . . . . . . . . . . . . . . . . . 29
3.3.8 Errors due to non-uniformity of collection of integrating spheres 30
3.3.9 Polarisation errors in the 0/45 geometry . . . . . . . . . . . . . 30

v
vi

3.3.10 Differences in methods of calculating color data from spectral data 30

3.3.11 Geometry differences between illumination and collection optics
within the specified limits . . . . . . . . . . . . . . . . . . . . . 31
3.3.12 Errors due to thermochromism in samples . . . . . . . . . . . . 31
3.3.13 Errors due to the dependence of spectral resolution on band-
width, scan speed and integration time . . . . . . . . . . . . . . 31
3.4 Measurements and corrections . . . . . . . . . . . . . . . . . . . . . . . 31
3.5 Intercomparison results . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.6 Determination of colorimetric uncertainties . . . . . . . . . . . . . . . . 44
3.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

4 Tristimulus integration 54
4.1 The ASTM weighting method . . . . . . . . . . . . . . . . . . . . . . . 55
4.2 Spectral bandpass error . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.3 Materials and methods . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

5 Thermochromism 65
5.1 Background of the study . . . . . . . . . . . . . . . . . . . . . . . . . . 65
5.2 Absorbance, transmittance, and optical density . . . . . . . . . . . . . 66
5.2.1 Transmitting samples . . . . . . . . . . . . . . . . . . . . . . . . 67
5.2.2 Opaque samples . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
5.3 Thermal effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
5.4 Numerical results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75
5.5 Thermochromic measurements . . . . . . . . . . . . . . . . . . . . . . . 76
5.6 Experimental results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
5.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79

6 Conclusions 82

References 84
 Chapter I
Introduction

Surface color measurement is important for a very wide range of industrial applications
including paint, paper, printing, photography, textiles, plastics [43,45]. Let us consider
an example. A manufacturer wants to produce goods with specific color appearance.
He can design the appearance of the product and he can measure it. After the produc-
tion process the appearance of the product might be changed or it can vary between
manufacturing batches because of noise inherent in the process. Now a manufacturer
would like to know if the actual color is sufficiently similar to the desired one or whether
the color is so different that it is not economically advisable to sell the products be-
cause of fear of consumer complaints. From this point of view, it is very important to
be able to handle measurements and calculations of small color differences [58].
For demanding color measurement, a spectral approach is definitely needed. Strictly
it is impossible to define the absolute color values of a sample, that is why we always
work with some kind of approximations, some of us closer than others. The human
eye can perceive color differences as small as 0.5 CIELAB units and, thus, distinguish
millions of colors. This 0.5 unit difference should be the goal for precise color measure-
ments. This limit is not a problem if we only want to measure the color difference of
two samples, but if we want to simultaneously know the exact color coordinate values
accuracy problems arise. The values of two instruments can be astonishingly different.
The best accuracy in color measurement could be achieved by use of a spectropho-
tometer. The only geometry for measuring real spectral reflectance is normal/diffuse
(0/d) geometry where the specimen is illuminated by a beam whose axis at an angle
which does not exceed 10◦ from the normal to the specimen [5, 60]. The reflected flux
should be collected by means of an integrating sphere.
The accuracy of the spectrophotometer used in color measurement may depend
on various errors such as photometric non-linearity, wavelength error, and integrating
sphere dark level error, integrating sphere error in both specular included and specular
excluded modes. Thus, correction formulas should be used to obtain more accurate

1
2 1. Introduction

results. Another question is how many channels i.e. wavelengths were used to measure
a spectrum. It is obvious that the sampling interval should be short enough to gain
more precise results. Furthermore, the result we attain is always a compromise between
measuring time, conditions and cost. Sometimes, one has to use a portable system or
the shape and the size of a sample makes it impossible to be able to use sensitive
equipment.
Regardless of how good an instrument one has, we need to point out that ultimately
color is a sensation produced in the human brain. It is evident that the information
processing system of the human has learned to process, e.g. visual information, in an
efficient way. The human eye is able to distinguish several million colors, and thus color
measuring instruments should accurately match the sensitivity of the eye to be able
to detect small color differences [18]. The basic sensory system of the eye is known,
but the operation of the process is an open question. The color of any non fluorescent
specimen can be matched with a mixture of red, green and blue primaries, because
there are only three types of color sensitive receptor on the retina. Human color vision
shows differences among people. Thus, some standard observers must be specified.
The CIE 1931 standard observer is defined for a 2◦ field of view by two equivalent sets
of color matching functions [5]. The first set is expressed in terms of spectral stimuli of
wavelengths 700 nm (R), 546.1 nm (G), and 435.8 nm (B). The second set is a linear
transformation of the first one and remains positive for all wavelengths. The CIE 1964
standard observer has been similarly defined for the 10◦ field of view. These sets are
used throughout industry.
As well as standard observers, standard illumination conditions must be defined as
the form of tables of relative spectral power against wavelength. Illuminant A corre-
sponds to the interior illumination by tungsten filament lamps, C represents average
daylight with a correlated color temperature of 6774 K, and the D sources aim to
describe for other phases of daylight. For instance, the most widely used daylight stan-
dard is D65 , which represents a source whose correlated color temperature is 6504 K.
There are a number of other illuminants which are used, too.
This thesis contains studies on accurate surface color measurements. First, some
basic background is introduced in Chapter II. Errors in a spectrophotometric measure-
ments are discussed and correction methods introduced in Chapter III. This work was
part of the European Union color measurement harmonisation project in 1997–2000
where 9 laboratories in Europe were involved. The author was responsible for measur-
ing, analysing and reporting the results made in University of Joensuu [15, 20, 43, 44].
In Chapter IV the wavelength interval in a tristimulus integration is considered. The
ASTM weighting functions are tested with a large data set of non fluorescent colors
and found to be useless. The results are first time reported in this thesis and will
be published later. The thermochromic effect is discussed in detail and measured in
Chapter V. The first idea of this phenomenon comes from the co-authors Jääskeläinen
 3

and Silfsten. However, formulaes introduced here, were derived by the author with co-
authors and entirely calculated numerically by the author [16, 19]. This thesis shows
how thermochromism is based on physical processes. Simple formulas are derived, and
shown to explain the experimental data. Such an explanation has not been published
in literature before.
In conclusion, this thesis shows how commercial instruments should be calibrated
for precision color measurements, if one aims at achieving the highest level for mea-
suring accuracy. In addition, the precise color measuring technique developed during
the preparation of this thesis, has been applied to a number of industrial projects.
Use of spectral data is increasing in a wide range of applications, too. The authors
contribution is shown in [17, 21–23, 30–32, 37, 38, 46–48] but the topic of this thesis was
the first phase: How to measure spectral data accurately.
 Chapter II
Colorimetry

The specification of basic standards used in colorimetry are based on definitions of

the Commission Internationale de l’clairage (CIE) by general consent in all countries.
The first major recommendations regarding colorimetric standards were made by the
CIE in 1931. The original recommendations made in 1931 are reviewed from time
to time by the CIE Colorimetry Committee and changes are made when considered
necessary. In this chapter, the CIE recommendations are briefly introduced since they
essentially form the basis of modern color research. The original and more precise
recommendations can be found from [5]. General and historical views of color vision
can be found from [3, 18, 57, 60].

2.1 Background to colorimetry

At the 6th Session of the CIE held at Geneva in 1924 it was decided to set up a Study
Group on Colorimetry (CIE 1924). The reason was the fact that the measurement of
color had become an important factor in industry and scientific laboratories but there
was not color specification system that could be considered satisfactory for general
practice. Later, it was agreed that efforts should be made to reach agreements on

• colorimetric nomenclature

• a standard daylight for colorimetry

• the “sensation curves” of the average human observer with normal color vision.

At the 8th Session of the CIE held at Cambridge, England, in 1931, the first major
recommendations were made which laid the basis for modern colorimetry (CIE 1931).
There was a total of five recommendations. Recommendations 1, 4 and 5 established
the CIE 1931 standard observer and a colorimetric coordinate system, recommendation

4
2.1 Background to colorimetry 5

2 specified three standard sources (A, B and C) and recommendation 3 standardized

the illuminating and viewing conditions for the measurement of reflective surfaces and
the standard of reflectance in the form of a magnesium-oxide surface.
The CIE 1931 standard colorimetric observer was defined by two different but
equivalent sets of color-matching functions based on the photopic luminous efficiency
function V (λ). These were already adopted by the CIE in 1924 and experimental
work was carried out by Guild [13] in 1931 and Wright [59] 1928–1929. The first
set of color-matching functions, r(λ), g(λ), b(λ), was expressed in terms of spectral
stimuli of wavelengths 700.0 nm (R), 546.1 nm (G), 435.8 nm (B) (Fig. 2.1), as the
reflectance stimuli, with the units adjusted so that the chromaticity coordinates of the
equi-energy spectrum are equal. The equi-energy spectrum is a stimulus whose spectral
concentration of power as a function of wavelength is constant.
0.4

b r
0.3
Tristimulus values

g
0.2

0.1

- 0.1
400 500 600 700
Wavelength [nm]

Figure 2.1: Color matching functions r, g and b in terms of spectral stimuli of wave-
lengths 700.0 nm (R), 546.1 nm (G) and 435.8 nm (B), respectively [60].

The second set of color-matching functions, x(λ), y(λ), z(λ) (Fig. 2.2), was rec-
ommended for reasons of more convenient application in practical colorimetry. Its
derivation from the first set was based on a proposal by Judd [24] in 1930 and involved
a linear transformation. The coefficients of the transformation were chosen so as to
avoid negative values of x(λ), y(λ), z(λ), at all wavelengths and so that the luminances
LX , LY , LZ of unit quantities of the stimuli were equal to 0, 1, 0 respectively, result-
ing in a set of color matching functions in which y(λ) is identical to V (λ). The units
of the new reference stimuli (X), (Y ), (Z) were adjusted to make the chromaticity
coordinates x, y, z also equal for the equi-energy spectrum.
6 2. Colorimetry

1.8
z

1.6

1.4

Tristimulus values
1.2
x
y
1

0.8

0.6

0.4

0.2

0
400 450 500 550 600 650 700 750
Wavelength [nm]

Figure 2.2: Color matching functions x, y and z [60].

The new color-matching functions, x10 (λ), y 10 (λ), z 10 (λ), defining the supplemen-
tary standard colorimetric observer (CIE 1959) were officially adopted in 1964. They
were derived from experimental data supplied by Stiles and Burch [51] in 1959 and
by Speranskaya [49] in 1959. The experimental color-matching data were obtained for
a 10◦ field by a direct method which did not involve an appeal to the CIE spectral
luminous efficiency function, but depend on the actual measurement of the relative
power distribution of the spectrum studied.
One of the important problems the Colorimetric Committee has dealt with over
several years concerns a coordinate system which would provide a three dimensional
color spacing that would be perceptually more uniform than the XY Z system. Many
different proposals have been forwarded over the years. In the 14th Session of the CIE
at Brussels in 1959 the committee considered a number of the systems. The MacAdam
uniform chromaticity scale diagram of 1937 was adopted as a standard UCS diagram.
The proposal was officially approved by the CIE in 1960, so the diagram is nowadays
known as the CIE 1960 UCS diagram.
At the 18th Session of the CIE in London in 1975 the Colorimetry Committee
approved the adoption of two new color spaces and associated color difference formulae.
These spaces are known as the CIE 1976 L∗ u∗ v ∗ color space and the CIE 1976 L∗ a∗ b∗
color space. The former was mainly used for TV and illumination industries and the
latter was used for surface color industries.
A lot of work has been done for decades to generate more uniform color spaces such
as CMC(l : c), BFD(l : c), CIE94, since the linearity of the CIE 1976 color spaces is not
satisfactory enough, at least where small color differences are concerned [4,26,36,41,42].
2.2 CIE Recommendations 7

Among the imaging industry there was still a need for a single color appearance model
that could be used throughout the industry to promote uniformity of practice and
compatibility between various components in a modern imaging system. The CIE was
able to respond these needs. In 1997 CIECAM97s color appearance model was adopted
by the CIE for color imaging applications [27–29, 33, 35].
For the color difference evaluation, a new color difference formula CIEDE2000 [34]
was developed, which is based on CIELAB space. It includes not only lightness, chroma,
and hue weighting functions but also an interactive term between chroma and hue dif-
ferences for improving the performance for blue colors and a scaling factor for CIELAB
a∗ scale for improving the performance for grey colors. It has been approved by the
CIE and published as a CIE Technical Report [6].
The present colorimetric recommendations were published in CIE Publication 15.2,
Colorimetry [5], in 1986. The new and up-to-date recommendations will be published
in 2002 as CIE Publication 15.31 .

2.2 CIE Recommendations

The terminology of the original recommendations has been altered to be consistent with
modern nomenclature and in some cases the original recommendations have also been
modified in contents to bring them into line with present day thinking and practice.
The recommendations are divided into the following five groups:

1. Recommendations concerning standard physical data.

2. Recommendations concerning standard observer data.

3. Recommendations concerning the calculation of tristimulus values and chromatic-

ity coordinates.

4. Recommendations concerning uniform color spacing.

5. Recommendations concerning miscellaneous colorimetric practices and formulae.

2.2.1 Standard physical data

Illuminants for colorimetry
Standard illuminants such as Illuminant A, B, C and D65 are recommended illuminants
for color calculations. Here “illuminant” refers to a specific spectral power distribution,
1
According to The 9th Congress of the International Colour Association, June 24–29, 2001,
Rochester, NY.
8 2. Colorimetry

not necessarily realized by a “source” which refers to a physical light emitter, such as
a lamp, the sun and the sky. Illuminant A represents light from a full radiator at
absolute temperature 2856 K according to The International Practical Temperature
Scale, 1968. The relative spectral power distribution of illuminant A has been derived
in accordance with Planck’s radiation formula.
Illuminant B was intended to represent direct sunlight with a correlated color tem-
perature of approximately 4900 K. Illuminant C was intended to represent average
daylight with a correlated color temperature of about 6800 K. In Fig. 2.3 standard
illuminants A, B and C are shown, where the radiant power has been normalized to the
same value at 550 nm. Illuminant D65 was intended to represent a phase of daylight
with a correlated color temperature of approximately 6500 K. The illuminant D65 is
recommended for use whenever possible. Illuminants D50 , D55 and D75 can still be used
to realize a phase of daylight having correlated color temperatures of approximately
5000 K, 5500 K, and 7500 K, respectively. In Fig. 2.4 examples of the standard day-
light illuminants D55 , D65 and D75 are shown. The radiant power has been normalized
to the same value at 550 nm.

A
Radiant power [a.u.]

B B

C
A

400 500 600 700 800

Wavelength [nm]

Figure 2.3: Standard illuminants A, B and C. The radiant power has been normalized
to the same value at 550 nm [60].

Sources representing illuminants

It is recommended that the following artificial sources are used to realize the illuminants
defined above. Standard illuminant A is to be realized by a gas-filled tungsten filament
2.2 CIE Recommendations 9

Radiant power [a.u.]

D 55
D 75
D 65
D 65
D 75
D 55

400 500 600 700 800

Wavelength [nm]

Figure 2.4: Standard illuminants D55 , D65 and D75 . The radiant power has been
normalized to the same value at 550 nm [60].

lamp operating at a correlated color temperature of 2856 K. A lamp with a fused-

quartz envelope or window is recommended if the spectral power distribution of the
ultraviolet radiation of illuminant A is to be realized more accurately.
Illuminant B and C are to be realized using the source A, combined with a filter
consisting of a layer of agreed solutions. At present, no artificial sources have been
recommended to realize illuminant D65 or any other D illuminants of various correlated
color temperatures.

Standard of the reflectance factor

The perfect reflecting diffuser is recommended as a reference standard in 1986 in publi-
cation CIE Standard Colorimetric Observers. It is defined as the ideal isotropic diffuser
with a reflectance equal to unity. Smoked magnesium oxide was superseded from Jan-
uary 1, 1969. A secondary reference standard, such as pressed barium sulphate, must
be calibrated in terms of the perfect reflecting diffuser.

Illuminating and viewing conditions for a reflecting specimen

CIE recommends the use of one of the following illuminating and viewing conditions:
45◦ /normal , normal/45◦ , diffuse/normal and normal/diffuse. For these conditions the
following symbols are used 45/0, 0/45, d/0 and 0/d, respectively.
In the 45/0 geometry the sample is illuminated by one or more beams whose effective
10 2. Colorimetry

axes are at the angle of 45◦ ± 2◦ from the normal to the sample surface. The viewing
angle from the normal to the sample should be less than 10◦ . There is also a restriction
in the viewing and illuminating beams: the angle between the axis and ray should not
exceed 8◦ .
In the 0/45 geometry the illumination is in the direction normal to the sample, and
the viewing angle is 45◦ from the normal. Now normal illumination is within 2◦ , and
the angle between the axis and any ray should not exceed 8◦ . The same restriction
should be observed in the viewing beam. The illuminatig conditions 0/45 and 45/0 are
shown in Fig. 2.5.

0/45 45/0

Figure 2.5: Viewing geometries of 0/45 and 45/0.

If the sample is illuminated diffusely by an integrating sphere and the viewing

angle does not exceed 10◦ from the direction normal to the surface of the sample,
one is referring to the d/0 geometry. The integrating sphere may be of any diameter
provided the total area of the ports do not exceed 10 percent of the internal reflecting
sphere area. The angle between the axis and arbitrary ray of the viewing beam should
not exceed 5◦ .
If the specimen is illuminated by a beam whose axis does not exceed 10◦ from the
normal to the specimen one is referring to 0/d geometry. The reflected flux is collected
by means of an integrating sphere. The angle between the axis and any ray of the
illuminating beam should not exceed 5◦ .
In Fig. 2.6, a typical setup of 0/d geometry is shown using an integrating sphere.
When integrating spheres are used, they should be equipped with white coated baffles
to prevent light passing directly between the sample and the spot of the sphere wall
illuminated or viewed.
2.2 CIE Recommendations 11

reference
beam

detector
sample
beam

gloss trap sample

or white baffles
sphere cap

reference

Figure 2.6: Integrating sphere.

In the 0/d and d/0 conditions, specular reflection can be excluded or included by
the use of a gloss trap. In the 0/d condition, the sample should not be measured with
a strictly normal axis of illumination if it is required to include the regular component
of reflection. Note that only the 0/d geometry provides a spectral reflectance. The
other conditions d/0, 0/45 and 45/0 give a specific radiance factor.

Illuminating and viewing conditions for transmitting specimens

It is recommended that the colorimetric specification of transmitting species corre-
spond to one of the following illuminating and viewing conditions: normal/normal,
normal/diffuse and diffuse/diffuse. The following symbols are used 0/0, 0/d and d/d,
respectively.
In the normal/normal (0/0) condition, the specimen is illuminated by a beam whose
effective axis is at a angle not exceeding 5◦ from the normal to its surface and with the
angle between the axis and any ray of the illuminating beam not exceeding 5◦ . The
geometric arrangement of the viewing beam is the same as that of the illuminating
beam. The specimen is positioned so that only the regularly transmitted flux reaches
the detector. This condition gives the regular transmittance, τr .
In the normal/diffuse (0/d) condition, the specimen is illuminated by a beam whose
effective axis is at an angle not exceeding 5◦ from the normal to its surface and with
the angle between the axis and any ray of the illuminating beam not exceeding 5◦ . The
hemispherical transmitted flux is usually measured with an integrating sphere. The
reflectance of the sphere reflecting surface or other material at the point of impingement
of the regularly transmitted beam, or at the point of impingement of the illuminating
12 2. Colorimetry

beam in the absence of a specimen, must be identical to the reflectance of the remainder
of the internal reflecting sphere area. This condition gives the total transmittance, τ .
If the regularly transmitted flux is excluded, for example by the use of a light trap, it
gives the diffuse transmittance, τ0/d . If the positions of the light source and detector
are interchanged, the method gives the equivalent diffuse/normal (d/0) quantities.
In the diffuse/diffuse (d/d) condition, the specimen is illuminated diffusely with an
integrating sphere and transmitted flux is collected using a second integrating sphere.
This condition gives the double transmittance, τdd .

2.2.2 Standard observer data

CIE 1931 standard colorimetric observer

It is recommended that colorimetric specification of color stimuli be based on the

color-matching functions x(λ), y(λ), z(λ) given in the CIE Standard on Colorimetric
Observers, whenever correlation with visual color matching of fields of angular subtense
between about 1◦ and about 4◦ at the eye of the observer is desired. These color-
matching functions are given in the Standard as values from 360 nm to 830 nm at
1 nm intervals with seven significant digits, and they define the CIE 1931 standard
colorimetric observer. The values at 5 nm intervals over the range 380 nm to 780 nm
are consistent with the Standard and are sufficient for most applications.

CIE 1964 supplementary standard colorimetric observer

It is recommended that colorimetric specification of color stimuli be based on the color-

matching functions x10 (λ), y 10 (λ), z 10 (λ) given in the CIE Standard on Colorimetric
Observers, whenever correlation with visual color matching of fields of angular subtense
greater than about 4◦ at the eye of the observer is desired. These color-matching
functions are given in the Standard as values from 360 nm to 830 nm at 1 nm intervals
with six significant digits, and they define the CIE 1964 standard colorimetric observer.

2.2.3 Calculation of tristimulus values and chromaticity coordinates

Calculation of tristimulus values

The CIE Standard on Colorimetric Observers recommends that the CIE tristimulus
values of a color stimulus be obtained by multiplying at each wavelength the value of
the color stimulus function φλ (λ) by that of each of the CIE color matching functions
and integrating each set of products over the wavelength range corresponding to the
entire visible spectrum 360 nm to 830 nm. The integration may be carried out by
2.2 CIE Recommendations 13

numerical summation at wavelength intervals, ∆λ, equal to 1 nm,

X = k φλ (λ) x (λ) ∆λ
λ


Y = k φλ (λ) y (λ) ∆λ (2.1)
λ


Z = k φλ (λ) z (λ) ∆λ,
λ

or

X10 = k10 φλ (λ) x10 (λ) ∆λ
λ


Y10 = k10 φλ (λ) y 10 (λ) ∆λ (2.2)
λ


Z10 = k10 φλ (λ) z 10 (λ) ∆λ,
λ

where X, Y , Z are tristimulus values, x(λ), y(λ), z(λ), are color-matching functions of
a standard colorimetric observer, and k is a normalizing constant defined below. These
equations may be written without or with the subscript 10 to correspond to the CIE
1931 or 1964 standard colorimetric system, respectively.
For reflecting or transmitting object colors, the color stimulus function, φλ (λ), is
replaced by the the relative color stimulus function, φ(λ), evaluated as

φ(λ) = R(λ)S(λ)
(2.3)
φ(λ) = τ (λ)S(λ),

where R(λ) is the spectral reflectance factor (or spectral radiance factor or spectral
reflectance) of the object color, τ (λ) is the spectral transmittance of the object color,
and S(λ) is the relative spectral power distribution of the illuminant. In this case, the
constants k and k10 are chosen so that Y = 100 for objects for which R(λ), or τ (λ) = 1
14 2. Colorimetry

for all wavelengths and hence

100
k = 
λ S(λ)y(λ)∆λ
(2.4)
100
k10 =  .
λ S(λ)y 10 (λ)∆λ

For self-luminous objects and illuminants the constants k and k10 are usually chosen
on the grounds of convenience. However, if in the CIE 1931 standard colorimetric
system the Y value is required to give the absolute value of photometric quantity the
constant k must be equal to Km , the maximum spectral luminous efficacy. This value
is equal to 683 lumens per watt and φλ (λ) must be the spectral concentration of the
radiometric quantity corresponding to the photometric quantity required.

The use of abridged or truncated data

The color stimulus function φλ (λ) should be known at 5 nm intervals over the wave-
length range from 380 nm to 780 nm. In practical applications, all the required data
may not be available. Data may have been measured at greater intervals than 5 nm or
it may not be equally divided. Many times it is possible to predict unmeasured data.
It is important to use the same wavelength interval and range throughout for any set
of precise color difference calculations.
Abridgement of the data may lead to errors in the computed tristimulus values.
Data with 10 nm or 20 nm intervals should be used only when it can be demonstrated
that these errors are negligibly small for the intended use of the tristimulus values.
If this is not the case it is recommended to interpolate needed but unmeasured val-
ues. Such a prediction should be made with a polynomial interpolation formula or by
Lagrange interpolation.
In some cases the measurement range is less than the practical range of summation
from 380 nm to 780 nm. Omission of the values at these limits of the measurement
range may lead to errors while computing tristimulus values. Such truncation should
be used only if it can be demonstrated that these errors are negligibly small. If these
errors are not negligibly small adequate extrapolation is recommended. The range
of the summation is an essential part of the tristimulus specification. As a rough
approximation, in the absence of other information, unmeasured values may be set
equal to the nearest measured value of the appropriate quantity in truncation.
2.2 CIE Recommendations 15

Calculation of chromaticity coordinates

The chromaticity coordinates (x, y, z) should be calculated from the tristimulus values
(X, Y, Z) as follows:

X
x =
X +Y +Z
Y
y = (2.5)
X +Y +Z
Z
z = .
X +Y +Z

Because of the relation x + y + z = 1, it is sufficient to quote x and y only. The

chromaticity coordinates x10 , y10 , z10 are computed similarly from the tristimulus values
X10 , Y10 , Z10 .

2.2.4 Uniform color spacing

The CIE 1976 UCS diagram
The CIE 1976 UCS chromaticity diagram is recommended for use whenever a pro-
jective transformation of the (x, y)-diagram yielding color spacing perceptually more
nearly uniform than that of the (x, y)-diagram is desired. The chromaticity diagram is
produced by plotting u

4X
u = (2.6)
X + 15Y + 3Z

as abscissa and v 
9Y
v = (2.7)
X + 15Y + 3Z

as ordinate, in which X, Y, Z are tristimulus values. The third chromaticity coordinate

w is equal to 1−u −v  . This diagram is intended to apply to comparisons of differences
between object colors of the same size and shape, viewed in identical white to middle-
grey surroundings, by an observer photopically adapted to a field of chromaticity not
too different from that of average daylight.
16 2. Colorimetry

The CIE 1976 uniform color spaces

The use of the following color spaces is recommended whenever a three-dimensional

spacing perceptually more nearly uniform than that provided by the XY Z system is
desired.
CIE 1976 L∗ u∗ v ∗ color space or CIELUV color space is defined by quantities
L ∗ , u∗ , v ∗

1/3
  
∗ Y Y
L = 116 − 16 when > 0.008856 (2.8)
Yn Yn
u∗ = 13L∗ (u − un ) (2.9)
v∗ = 13L∗ (v  − vn ), (2.10)

where Y, u , v  describe the color stimulus considered and Yn , un , vn describe a specified
white object color stimulus. If Y /Yn is less than 0.008856, the above equations are
changed as follows

 1/3  
Y Y 16
is replaced by 7.787 + .
Yn Yn 116

The CIE 1976 L∗ a∗ b∗ color space or CIELAB color space is defined by quantities
L∗ a∗ b ∗

 1/3
∗ Y
L = 116 − 16 (2.11)
Yn
   1/3 
1/3
X Y
a∗ = 500 − (2.12)
Xn Yn
   1/3 
1/3
Y Z
b∗ = 200 − , (2.13)
Yn Zn

where X, Y, Z describe the color stimulus considered and Xn , Yn , Zn describe a specific

white object color stimulus. If X/Xn , Y /Yn or Z/Zn is less than 0.008856, the above
2.2 CIE Recommendations 17

equations are changed as follows

 1/3  
X X 16
is replaced by 7.787 +
Xn Xn 116
 1/3  
Y Y 16
is replaced by 7.787 +
Yn Yn 116
 1/3  
Z Z 16
is replaced by 7.787 + .
Zn Zn 116

∗ ∗
The differences ∆Euv or ∆Eab between two color stimuli are calculated as the
Euclidean distance between the points representing them in the space:

∗

∆Euv = (∆L∗ )2 + (∆u∗ )2 + (∆v ∗ )2 (2.14)

∗

∆Eab = (∆L∗ )2 + (∆a∗ )2 + (∆b∗ )2 . (2.15)

2.2.5 Miscellaneous colorimetric practices and formulae

Dominant wavelength
The dominant wavelength of a color stimulus λd is a wavelength of the monochromatic
stimulus that matches the color stimulus considered when additively mixed in suit-
able proportions with the specified achromatic stimulus. A monochromatic stimulus
is monochromatic radiant power of given magnitude and wavelength, entering the eye
and producing a sensation of light or color. An achromatic stimulus is the color stim-
ulus chosen because it usually yields a color perception which is devoid of hue under
the desired observing conditions. Complementary wavelength is used instead of dom-
inant wavelength for stimuli whose chromaticities lie between those of the specified
achromatic stimulus and the purple line.

Complementary wavelength
The complementary wavelength of a color stimulus λc is a wavelength of the monochro-
matic stimulus that matches the specific achromatic stimulus when additively mixed
in suitable proportions with the color stimulus considered.
18 2. Colorimetry

Colorimetric purity
The colorimetric purity pc is defined by the relation

Ld
pc = , (2.16)
Ln + Ld

where Ld and Ln are the luminances of the monochromatic stimulus and of the speci-
fied achromatic stimulus that match the color stimulus considered in an additive mix-
ture, respectively. In the case of stimuli characterized by a complementary wavelength
suitable mixtures of light from the two end of the spectrum are used instead of the
monochromatic stimuli. In the CIE 1931 standard colorimetric system, colorimetric
purity is related to excitation purity pe by the equation

yd
pc = pe , (2.17)
y

where yd and y are the y-chromaticity coordinates of the considered monochromatic

stimulus and the color stimulus, respectively.

Excitation purity
Excitation purity pe is defined by the ratio N C/N D of two collinear distances on the
chromaticity diagram of the CIE 1931 or 1964 standard colorimetric system. The first
distance is that between point C and N which represents the considered color stimulus
and the specified achromatic stimulus, respectively. The second distance is between
point N and point D on the spectrum locus at the considered dominant wavelength of
the color stimulus. The definition leads to the following expressions

y − yn x − xn
pe = or pe = , (2.18)
yd − yn xd − xn

where (x, y), (xn , yn ), (xd , yd ) are the x, y-chromaticity coordinates of the points C, N
and D, respectively.

Special metamerism index

Two specimens having identical tristimulus values for a given reference illuminant and
reference observer are metameric if their spectral radiance distributions differ within
2.2 CIE Recommendations 19

the visible spectrum. It is recommended that two specimens whose corresponding

tristimulus values (X1 = X2 , Y1 = Y2 , Z1 = Z2 ) are identical with respect to a reference
illuminant and observer, the metamerism index, Mt is set to be equal to the index
of color difference ∆E between the two specimens computed for the test illuminant
t. The preferred reference illuminant is the CIE standard illuminant D65 . If another
illuminant is used as reference this should be noted.

The evaluation of whiteness

According to CIE recommendations, the formulae W or W10 for whiteness and TW
or TW,10 for tint, given below, are used for comparisons of the whiteness of samples
evaluated for CIE standard illuminant D65 . The application of the formulae is restricted
to samples that are called “white” commercially, that do not differ much in color and
fluorescence, and that are measured on the same instrument at nearly the same time.

W = Y + 800(xn − x) + 1700(yn − y) (2.19)

W10 = Y10 + 800(xn,10 − x10 ) + 1700(yn,10 − y10 ) (2.20)

TW = 1000(xn − x) − 650(yn − y) (2.21)

TW,10 = 900(xn,10 − x10 ) − 650(yn,10 − y10 ), (2.22)

where Y is the Y -stimulus value of the sample, x and y are the x, y chromaticity
coordinates of the sample, and xn , yn are the chromaticity coordinates of the perfect
diffuser, all for the CIE 1931 standard colorimetric observer. Y10 , x10 , y10 , xn,10 and yn,10 ,
are similar values for the CIE 1964 supplementary standard colorimetric observer. The
higher the value of W or w10 the greater is the indicated whiteness. The more positive
the value TW or TW,10 the greater is the indicated greenishness. The more negative the
value TW or TW,10 the greater is the indicated reddishness. For perfect diffuser W and
W10 are equal to 100 and TW and TW,10 are equal to zero.

Calculation of correlated color temperature

The correlated color temperature of a given stimulus is the temperature of the Planck-
ian radiator whose perceived color most closely resembles that of the stimulus at the
same brightness and under the same viewing conditions. The recommended method
for calculating the correlated color temperature of a stimulus is to determine on a chro-
maticity diagram the temperature corresponding to the point on the Planckian locus
that is intersected by the agreed isotemperature line containing the point representing
the stimulus.
20 2. Colorimetry

2.3 Summary
The recommendations of the CIE Colorimetry Committee are given for various col-
orimetric practices and formulae such as recommendations for standard illuminants,
for the standard reflectance factor, for illuminating and viewing conditions, for the
standard colorimetric observers, for the calculation of tristimulus values, chromaticity
coordinates, and color differences.
 Chapter III
Measurement of surface color

Surface color measurements are of great importance in the production of a very wide
range of manufactured goods in industry. Industry needs to be able to measure surface
∗
color to within the discrimination limits of the human eye of 0.5 ∆Eab CIELAB units.
The basic idea is to measure the absolute spectrum as accurately as possible [25].
The spectrum is then used with the stored numerical data to derive desired results.
The algorithms should include the commonly used illuminants together with various
coordinate systems, whiteness formulas, and color difference formulas.
There are two main classes of color measuring instruments, which are used for
surface color measurements, colorimeters and spectrophotometers. Colorimeters are
trichromatic devices, where the illuminant is simulated by a light source and the color
matching functions are simulated by (interference) colored filters in combination with
a photodetector. These instruments have a simple construction, but they are not
accurate enough for typical quality control tasks in industry.
Spectrophotometers allow accurate measurement, because they measure the spec-
tral reflectances of the samples. The disadvantage of traditional scanning spectropho-
tometers is that they are slow. In addition, because of vibration sensitivity and other
environmental requirements they can not be used outside of laboratory conditions.
These disadvantages have recently been avoided by constructing non-scanning spec-
trophotometers, which are based on use of diode array detectors.

3.1 General considerations on color measurement

Internationally accepted illumination and observation angles should be standardised
for use in instruments such as normal/diffuse with including or excluding the specular
component, or 0/45. Tolerances on these angles along with the maximum angle of
acceptance are given in CIE Publication 15.2 [5].
If wide spectrum illumination is used (incandescence lamp, xenon lamp, flash lamp),
one should be aware of fluorescent samples. The source has to match a CIE standard or

21
22 3. Measurement of surface color

recommended illuminant (A, D65 , C etc.) as closely as possible to correctly determine

color coordinates. If this is not accomplished, the surface color of a fluorescent sample
has to be measured by the double monochromator method [40, 44]. Fluorescence is
present in many objects, even in some white standards.
There are some general principles to follow in order to obtain reliable measurement
results. Color measurement instruments have to be maintained as advised in their
manuals. It is necessary to pay special attention to the handling and maintance of the
instrument’s calibration standards. The same attention should be paid to integrating
sphere coatings as well. Degradation of this coating produces several errors.
The instrument’s start up routine should be done as indicated by the manufacturer.
Currently this is automatically done by most spectrophotometers and colorimeters.
The warm-up time should also be followed as indicated in the manual. At the very
least the warm up should be thirty minutes in every case. Samples should be measured
at the same room temperature as the instrument.
Many instruments have their own calibration set, which contains a white standard
and a black standard or zero reflectance standard. The low reflectance standard is not
as common as the white one. However, it is advisable to use another calibration set to
independently check the instrument’s performance. Ideally, this set should contain the
following elements: a surface mirror, a wavelength standard, two white standards (one
matt and one glossy), a zero reflectance standard and at least one grey standard with
about 50% reflectance. The function of these elements is given later in this chapter.
It is good policy to measure a sample a few times and, to calculate the mean
value of those measurements. One can rotate the sample between each measurement
to avoid directional dependency from surface inhomogeneties. For as accurate results
as possible, the performance and potential error sources in the instrument must be
known [7, 8, 15, 20, 43].

3.2 Intercomparison
The human eye can perceive color differences approximately 0.5 ∆Eab in CIELAB
units. The first European intercomparison stopped in 1993 [45]. This intercomparison
sought to determine the state of the art of measurements of spectral reflectance and
color specification of surface colors using spectrophotometry. Four laboratories, each
from a different country within the European Community, participated. Sets of four-
teen ceramic color standards were calibrated by each laboratory. The intercomparison
covered the specular excluded, specular included and 0/45 geometries. The result was
that approximately half of the measurements made by the laboratories with respon-
sibility for national standards did not agree on the limit of the human eye. This is
inadequate for industrial requirements.
The second European intercomparison was started in 1997 [43]. The main objective
3.2 Intercomparison 23

in this project was to harmonise color measurements between 8 national laboratories to

achieve a target of 95% agreement within 0.5∆Eab CIELAB units for non-fluorescent
colors. This was achieved through the use of a system for determining and correcting
errors, followed by an intercomparison of non-fluorescent surface color measurements.
The second objective was to extend the capability for measurement of fluorescent safety
colors using the two monochromator method to three countries within the European
Union, and carry out an intercomparison of fluorescent colored materials to meet the
metrological requirements of EN471, Specification of High Visibility Warning Cloth-
ing. Partners in this project entitled “Harmonisation of National Scales of Surface
Color Measurements”, Contract SMT4-CT96-2140, European Commission Standards
Measurements and Testing Programme, were as follows:

• National Physical Laboratory (NPL), United Kingdom.

• Consejo Superior de Investigatines Cientificas (CSIC), Spain.

• Bundesanstalt für Materialforchung und-Prüfung (BAM), Deutchland.

• University of Joensuu (UJ), Finland.

• Laboratoire National d’Essais (LNE), France.

• Centro de Ciencias e Technologias Opticas (CETO), Portugal.

• Swedish National Testing Research Institute (SP), Sweden.

• British Ceramic Researc Ltd (BCRA), United Kingdom.

• Danish Electronic Light and Acoustics (DELTA), Denmark.

NPL worked as a co-ordinator of the project. During the first year, all partners
met at NPL. They agreed upon which errors in the measurements of surface color
that could be determined and corrected, and the principles of the methodology for
achieving this. NPL calibrated and dispatched to partners a set of calibrated artefacts
for determining correcting errors. Partners then determined the errors within their
instruments and reported their result to the co-ordinator. NPL also measured sets of 4
color standards to be used by partners for testing the effectiveness of the methodology
for harmonisation, and to give preliminary results on levels of agreement that might
be achieved for non-fluorescent colors.
24 3. Measurement of surface color

3.3 Errors in surface color measurements

Spectrophotometric errors are discussed in this section on a qualitative basis. In order
to achieve the target level of agreement of 0.5 ∆Eab CIELAB units, it is necessary
to determine and correct errors in the measurement color and to adopt a common
procedure where relevant. The following is a list of the principal errors:

• Errors in absolute scales of diffuse reflectance and 0/45 radiance factor.

• Errors due to differing properties of white reference standards.

• Non-linearity of the photodetector.

• Incorrect zero level.

• Wavelength scale error.

• Specular beam exclusion error.

• Specular beam weighting error.

• Errors due to non-uniformity of collection of integrating spheres.

• Polarisation errors in the 0/45 geometry.

• Differences in methods for calculating color data from spectral data.

• Errors due to thermochromism in samples.

• Errors due to the dependence of spectral resolution on bandwidth, scan speed

and integration time.

• Geometry difference between illumination and collection optics within the speci-
fied limits.

At present, there is no method for quantifying the effects of geometry differences

(last item) and applying corrections. The method used to minimise error due to
non-uniformity of collection by integrating spheres, which is to measure matt sam-
ples against matt masters and glossy samples against glossy masters, has not proved
very effective. Thus, it is now clear that there are significant undetermined errors
which remain uncorrected, limiting agreement at present.
Integrating sphere errors are due to the fact that the integrating sphere is not an
ideal sphere but a hemispherical sphere. There are also some baffles inside the sphere
which prevent straight light from striking the detector.
3.3 Errors in surface color measurements 25

3.3.1 Errors in absolute scales of diffuse reflectance and 0/45 radiance

factor

The method used for this intercomparison relies on the stability of the white reference
tiles to accurately transfer common absolute scales to all participants with a high degree
of accuracy. All standards were calibrated by NPL, providing a common scale for the
matt and glossy calibrations. These scales contain an inherent uncertainty associated
with the NPL calibration.

3.3.2 Errors due to differing properties of white reference standards

When making measurements of matt samples using spectrophotometers with a inte-

grating sphere accessory, errors may be introduced if the instrument is calibrated using
a glossy standard. The same is also true for glossy measurement against a matt stan-
dard. To minimize these errors a matt standard should be used for matt samples and
a glossy standard for glossy samples.

3.3.3 Photometric non-linearity

A lack of linearity in a detection system causes errors in results. Photometric non-

linearity error assumes that the response of the photodetector is not linear over the
reflectance scale from 0% to 100%. Towards the higher end of the signal range, satura-
tion of the detector or electronics may occur. There may also be some additional effects
such as inter-reflection between glass surfaces within the instrument and a reduction
in sphere response with dark samples.
The simplest type of relationship between the true value of reflectance R and the
instrumental value R is a quadratic form. That means the difference is biggest at the
50% reflectance level. Now the photometric non-linearity can be corrected by measuring
only one grey tile, about 50% of the reflectance. Definition of the non-linearity error
is based on the assumption that the error is zero at 100% and 0%. The 100% level
corresponds to a sample being measured against itself and 0% level errors are classified
as the dark level error and, thus, treated separately.
Assuming the quadratic relationship for the error we arrive at

R = R + δR
= a + bR + cR2 . (3.1)

By this definition we find that when R = 0 also R = 0 and a = 0. In the same way
26 3. Measurement of surface color

when R = 100, R = 100 also. Now we have the equations

100 = 100(b + 100c) (3.2)

b = 1 − 100c. (3.3)

For a grey tile, we can denote R = r and R = r . Thus

r = br + cr2
= (1 − 100c)r + cr2
= r − 100cr + cr2 . (3.4)

For a terms of c and b, we can solve

r − r
c = (3.5)
100r − r2

b = 1 − 100c
100r − 100r
= 1−
100r − r2
100r + r2
= . (3.6)
100r − r2

Hence,
   
100r − r2  r − r
R= R + R2 , (3.7)
100r − r2 100r − r2

where r is the certified value for the grey tile, and r is the corresponding measured
value. Now the linearity corrected spectrum R can be calculated for any color when
the uncorrected spectra R is measured.
As the tiles used in this intercomparison were calibrated against a white master on
a sphere instrument at NPL, it is possible that the NPL’s instrument’s non-linearity
may have been imposed on all values.
3.3 Errors in surface color measurements 27

3.3.4 Incorrect zero level

An integrating sphere dark level error is mainly due to light scattered inside the optics
of the spectrophotometer. This effect gives a halo around the main light beam, some
of which falls on the sphere wall. There may also be a component due to an electronic
offset. The error is determined by placing a black glass wedge at the sample port, which
absorbs the light in the main beam and by measuring the variation of reflectance reading
with a wavelength. An optical wedge gives a reading ρk . Absolute diffuse reflectance
of a sample Ra is

ρ a − ρk
Ra = Rt
ρ t − ρk

ρa 1 − ρρka
=  Rt
ρa 1 − ρρkt
  
ρa ρk ρk
≈ 1− 1+ Rt
ρt ρa ρt
 
ρa ρ k ρk
≈ 1− + Rt
ρt ρa ρt
 
ρa ρa
= Rt + k1 −1 , (3.8)
ρt ρt
where,
ρk
k1 = Rt . (3.9)
ρt

ρa in Eq. 3.8 is the sample reading, ρt is the white standard reading, ρk is the dark
reading and Rt is the calibrated value for a white standard.

3.3.5 Wavelength scale error

Wavelength error will lead to color errors, so the instrument’s wavelength scale should
be checked. Reflectance values should be corrected, according to the expression

∂Rm (λ)
R(λ) = Rm (λ) + ∆λ, (3.10)
∂λ

where R(λ) is the correct reflectance value, Rm is the measured reflectance value and
∆λ is the wavelength scale error, the difference between the actual and displayed value.
28 3. Measurement of surface color

Wavelength error can be determined by using several wavelength standards: spectral

lamps, reflection tiles or transmission filters. The user must use the most appropriate
standard for the instrument type.

3.3.6 Specular beam exclusion error

For a glossy sample measured in the specular excluded mode, the specular beam falls
on a gloss trap. Ideally, none of the light entering the trap should be reflected back into
the sphere and none of the beam should meet an integrating sphere wall. In practice,
these ideal conditions are not often achieved and, thus, correction is needed. The
simplest method is to use a calibrated front surface mirror to give a strong specular
beam with the white tile used as the diffuse reflectance standard. A calibrated mirror
of reflectance M gives a reading ρgm . In a case where a gloss trap is used to block
specular reflectance, a calibrated mirror M will give a reading

4
ρg4 = ρgm . (3.11)
M

The number 4 is used as a nominal value for the specular reflectance of a typical glossy
sample. Absolute diffuse reflectance Ra of a sample is

ρa − αρg4
Ra = Rt , (3.12)
ρt − βρg4

where α, β = 0 for a matt surface and α, β = 1 for a glossy surface. Using the same
approximates as for the dark error correction

 
ρa ρa
Ra ≈ Rt + k2 β − α , (3.13)
ρt ρt

where,
ρg4
k2 = Rt . (3.14)
ρt

Here with the ρa reading of a sample, ρt is read when the white standard is measured,
and Rt is the calibrated value for a white standard.
3.3 Errors in surface color measurements 29

3.3.7 Specular beam weighting error

For a glossy sample measured in the specular included mode, non uniformity of the
integrating sphere may mean that the specular component is not collected with the
same efficiency as the diffusely reflected light. A calibrated mirror of reflectance M
placed at the sample port gives a reading ρsm . If there is no specular beam error, the
mirror would give a reading

M
ρt , (3.15)
Rt

where ρt is a reading of a calibrated matt white tile Rt is placed at the sample port.
Thus, the error in the mirror reading em is given by

M
em = ρsm − ρt . (3.16)
Rt

If we apply the nominal value of 4% specular reflectance from a typical glossy sample
we arrive at the error in the specular beam

 
4 M
es = ρsm − ρt . (3.17)
M Rt

The absolute reflectance Ra of a sample when applying a correction is

ρa − αes
Ra = Rt
ρt − βes
 
ρa αes βes
≈ 1− + Rt
ρt ρa ρt
 
ρa ρa β α
= Rt + es − Rt , (3.18)
ρt ρt ρt ρa
30 3. Measurement of surface color

where α, β = 0 is for a matt surface and α, β = 1 is for a glossy surface, where the
same approximates are used as before. Substituting es into the above equation we get

  
ρa ρa 4 M β α
Ra = Rt + ρsm − ρt − Rt
ρt ρt M Rt ρt ρa
  
ρa ρsm 1 1 ρa
= Rt + 4 − β − α Rt
ρt ρ M Rt ρt
 t  
ρa ρsm Rt ρa
= Rt + 4 −1 β −α
ρt ρt M ρt
 
ρa ρa
= Rt + k3 β − α , (3.19)
ρt ρt

where,
 
ρsm Rt
k3 = 4 −1 . (3.20)
ρt M

In Eqs. 3.19 and 3.20 ρa is a sample reading, ρsm is a reading for a calibrated mirror
in a sample port, ρt is a reading for the calibrated white standard, M is the calibrated
value for a mirror and Rt is the calibrated value for a white standard.

3.3.8 Errors due to non-uniformity of collection of integrating spheres

The internal reflectance of an integrating sphere will usually vary over its surface. In
addition, baffles included within the sphere lead to a variation in response to the angle
of reflectance. The reflectance of samples, particularly materials such as metals and
paper, varies considerably with the direction of view. This, linked with the non-uniform
angular response of the sphere, may lead to errors.

3.3.9 Polarisation errors in the 0/45 geometry

Several researchers have carried out studies on the effects of polarisation. However, no
need for a correction has arisen in typical surface color samples.

3.3.10 Differences in methods of calculating color data from spectral data

In the calculation of colorimetric values some errors may occur. Usually, the CIE
standard observer data are available at 5 nm intervals. Errors may occur if calculation
is made at different intervals.
3.4 Measurements and corrections 31

3.3.11 Geometry differences between illumination and collection optics

within the specified limits
The tolerances given in the CIE geometry specification are sufficiently large to result
in a wide range of implementation in instrumental design.

3.3.12 Errors due to thermochromism in samples

In common with all materials, the ceramic tiles used in the intercomparison will change
color with temperature. Thus, samples should be measured at the same temperature.
It was agreed to measure samples in 23 ± 1◦ C in this comparison. In this thesis,
thermochromism is discussed in more detail in Chapter V.

3.3.13 Errors due to the dependence of spectral resolution on bandwidth,

scan speed and integration time
A fast speed scan with slow integration time will generate a different result to a slow
speed, fast integration time scan, particularly in the resolution of slopes. The difference
between the two resulting spectra might appear as a wavelength shift and particularly
affect the results for the more chromatic tiles. The same is true for an instrument with
a differently shaped spectral bandwidth. The partners were investigating the effects of
the scan speed and integration time and finally agreed to use same kind of parameters.

3.4 Measurements and corrections

A double beam scanning spectrophotometer, Perkin Elmer λ-18, was used in reflectance
spectra recording. Measurements were performed with a bandwidth of 2 nm and with a
scanning speed of 240 nm/min. The minimum temperature during the measurements
was 22.2◦ C while the maximum was 23.4◦ C. All samples were measured 6 times, 3
times in the morning and 3 times in the afternoon after at least a 3 hour interval. No
significant differences were noticed between the results. Error correction factors k1 , k2
and k3 were determined as described earlier in Eqs 3.9, 3.14 and 3.20. Quantities of
constants ki for a PE-18 instrument are shown in Fig. 3.1 as a function of wavelength.
The correction constants were then used to apply the needed corrections. Wavelength
error was determined by measuring the emission spectrum of a deuterium lamp. This
error was noticeably negligible. The amount of linearity correction was also determined
by measuring the mid-grey calibration sample which was compared to the calibrated
values. The difference between the calibrated and the measured values was about
0.56%-unit. As a matter of fact, the influence of the photometric linearity correction
was biggest among all applied corrections in our case. One has to point out that by
using only one grey tile a quadratic relationship is assumed for the linearity error. To
ensure the nature of the behavior of the photodetector, more calibrated grey samples
32 3. Measurement of surface color

should be used. One interesting finding during this intercomparison was that a grey
tile and a neutral transmittance filter, at about the same level of 50% refelectance, do
not necessarily have the same amount of non-linearity error. This was confirmed by
measurement between UJ and NPL. Later, in 1998 the co-ordinator received a letter
from Japan as part of the work of a CIE Technical Committee, giving completely
independent results.
0.06

0.05

0.04
Value

0.03

0.02

0.01

0
350 400 450 500 550 600 650 700 750 800
Wavelength [nm]

Figure 3.1: Error correction factors for the PE-18 spectrophotometer. Solid, dashed,
and dot-dashed lines correspond to the correction factors k1 , k2 and k3 , respectively.

3.5 Intercomparison results

The main objective of the project was to harmonise color measurements between 8
national laboratories within discrimination limits of the human eye. A target was
set to 95% agreement within 0.5 ∆E CIELAB units. Each laboratory measured and
applied corrections for a set of 16 samples, 8 matt samples and 8 glossy samples. All
of the measurements and calculations of Joensuu cite were done by the author. All
the samples were ceramic tiles which are known to be very stable [12]. The matt
samples were measured against a matt white tile and, similarly, glossy tiles against a
glossy white tile. Fig. 3.2 shows the reflectance spectra of the 8 matt color samples.
The colors of the tiles were pale grey, mid grey, black, red, bright yellow, green, cyan
and deep blue. These samples were provided and calibrated by the NPL. Partners
did not have a priori knowledge of the spectral data of the samples while making
measurements. Each laboratory tried to measure samples as accurately as they could.
3.5 Intercomparison results 33

Matt samples were measured in specular included geometry and glossy samples were
measured both in specular included and excluded geometry. Some of the laboratories
measured the samples in 0/45 geometry, as well.
100

90

80

70
Reflectance [%]

60

50

40

30

20

10

0
400 450 500 550 600 650 700 750
Wavelength [nm]

Figure 3.2: Reflectance spectra of a 8 matt color samples.

Part of the results of the comparison are shown in the Tables 3.1–3.17. Table 3.1
show the effect of the applied error corrections for the data measured at the University
of Joensuu. In specular included geometry corrections are made in the following order:
dark error correction (Rd ), specular beam error correction (Rs ) and non-linearity cor-
rection (Rnl ). Each correction was added cumulatively to the previous corrections. In
the specular excluded geometry a gloss trap error correction (Rg ) was applied instead
of the specular beam correction.
The colormetric differences from the NPL-data were calculated for each partner.
The results for the specular included geometry are shown in Tables 3.2 and 3.3 before
and after the corrections, respectively. In Tables 3.4 and 3.5 corresponding results are
shown for the specular excluded geometry. By comparing the results between 3.2 and
3.3 or 3.4 and 3.5 one can see the impact of applied corrections. In most corrections the
results improve. Surprisingly, there are quite large differences with black, red and deep
blue samples. It is hard to say whether the reason lies in the instrument or if there
is human error in the measurements. Nevertheless, the UJ-results are the only ones
where the impact of error corrections did not change results in the wrong direction, i.e.
in every case better results were achieved after corrections were applied. This indicates
that NPL and us were the best among the participating laboratories. However, we
need to point out that in this kind of experimental investigation the “right” results do
not exist. It is possible that there may be some error in the NPL-data as well. In that
34 3. Measurement of surface color

Table 3.1: CIELAB Colorimetric differences ∆E for the UJ-data after applied cumu-
lative corrections in the order of dark error (Rd ) correction, specular beam error (Rs )
or gloss trap error (Rg ) correction and non-linearity error (Rnl ) correction, 10 degree
observer, illuminant D65 .

Specular Included Specular Excluded

Geometry Geometry
Tile Rd Rs Rnl Rd Rg Rnl

Glossy Grey 0.00 0.01 0.23 0.00 0.00 0.25

Glossy Mid Grey 0.01 0.02 0.33 0.01 0.00 0.35
Glossy Black 0.03 0.08 0.20 0.11 0.14 0.21
Glossy Red 0.03 0.02 0.27 0.12 0.26 0.31
Glossy Bright Yellow 0.03 0.02 0.31 0.04 0.14 0.33
Glossy Green 0.01 0.03 0.35 0.02 0.07 0.37
Glossy Cyan 0.01 0.02 0.34 0.01 0.06 0.37
Glossy Deep Blue 0.04 0.15 0.36 0.15 0.90 0.87
Matt Grey 0.00 0.23
Matt Mid Grey 0.01 0.34
Matt Black 0.03 0.25
Matt Red 0.02 0.29
Matt Bright Yellow 0.02 0.29
Matt Green 0.01 0.33
Matt Cyan 0.01 0.32
Matt Deep Blue 0.03 0.27
3.5 Intercomparison results 35

case there will be some uncertainty in all error corrections. From this aspect one can
not directly conclude that smaller differences are always better.
First of all, the purpose of this intercomparison was to harmonise measurement
results between all partners. This means that differences are calculated from the mean
value of all measurements. Tables 3.6–3.9 show the percentage of the measurements of
each partner below three different ∆E limits from the mean. However, there may be
some minor disadvantages in this measure as well, since there are some large differences
among some colors. This will set the mean value as slightly false and distort the results.
36 3. Measurement of surface color

Table 3.2: Partners’ colorimetric differences ∆E from NPL for the set of 16 color tiles,
specular included geometry. Uncorrected results, 10 degree observer, illuminant D65 [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA

Glossy

Pale Grey 0.23 0.31 0.09 0.19 0.15 0.27 0.02 0.21
Mid Grey 0.37 0.41 0.59 0.14 0.28 0.34 0.13 0.59
Black 0.17 0.33 4.06 0.33 1.27 0.32 0.78 1.46
Red 0.73 0.69 4.04 0.68 1.45 0.70 0.92 1.58
Bright Yellow 0.59 0.63 3.25 0.46 1.19 0.66 0.63 1.48
Green 0.34 0.49 1.70 0.33 0.73 0.43 0.63 0.72
Cyan 0.47 0.51 1.37 0.57 0.50 0.49 0.80 0.51
Deep Blue 0.20 0.27 4.47 0.44 1.14 0.29 0.45 1.78

Matt

Grey 0.12 0.11 0.19 0.09 0.22 0.10 0.08 0.26

Mid Grey 0.43 0.20 0.09 0.25 0.06 0.11 0.26 0.37
Black 0.43 0.16 1.76 0.23 0.12 0.35 0.11 0.37
Red 0.71 0.78 1.01 0.56 2.55 0.62 0.79 1.30
Bright Yellow 0.84 0.47 1.56 1.31 0.71 0.76 0.82 0.96
Green 0.49 0.18 0.75 0.40 0.32 0.47 0.47 0.62
Cyan 0.42 0.39 0.70 0.48 0.22 0.40 0.73 0.31
Deep Blue 0.55 0.21 2.16 0.37 0.48 0.46 0.22 0.78

Average
|difference| 0.44 0.38 1.74 0.43 0.71 0.42 0.49 0.83
3.5 Intercomparison results 37

Table 3.3: Partners’ colorimetric differences ∆E from NPL for the set of 16 color
tiles, specular included geometry. Fully corrected results, 10 degree observer, illuminant
D65 [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA

Glossy

Pale Grey 0.05 0.31 0.12 0.18 0.20 0.05 0.08 0.24
Mid Grey 0.05 0.41 0.37 0.21 0.24 0.07 0.06 0.40
Black 0.17 0.33 2.72 0.33 0.57 0.17 0.20 0.28
Red 0.72 0.69 2.67 0.42 0.89 0.55 0.82 1.48
Bright Yellow 0.33 0.63 1.91 0.67 0.64 0.34 1.01 1.18
Green 0.26 0.49 1.08 0.50 0.44 0.40 0.55 0.95
Cyan 0.51 0.51 0.97 0.63 0.36 0.52 0.81 0.48
Deep Blue 0.42 0.27 3.06 0.23 0.34 0.14 0.33 0.36

Matt

Pale Grey 0.13 0.11 0.33 0.09 0.13 0.14 0.06 0.20
Mid Grey 0.10 0.20 0.19 0.28 0.10 0.29 0.10 0.47
Black 0.16 0.16 1.64 0.21 0.23 0.05 0.08 0.47
Red 0.57 0.78 1.84 0.46 2.49 0.46 0.80 1.33
Bright Yellow 0.59 0.47 1.34 1.41 0.78 0.47 0.84 1.04
Green 0.33 0.18 0.70 0.46 0.31 0.42 0.43 0.62
Cyan 0.39 0.39 0.75 0.54 0.19 0.39 0.67 0.21
Deep Blue 0.39 0.21 2.17 0.32 0.49 0.27 0.15 0.77

Average
|difference| 0.32 0.38 1.30 0.43 0.52 0.30 0.44 0.65
38 3. Measurement of surface color

Table 3.4: Partners’ colorimetric differences ∆E from NPL for the set of 8 color tiles,
specular excluded geometry. Uncorrected results, 10 degree observer, illuminant D65 [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA

Glossy

Pale Grey 0.24 0.15 0.08 0.20 0.16 0.28 0.02 0.13
Mid Grey 0.51 0.20 0.22 0.22 0.18 0.44 0.06 0.07
Black 0.60 1.17 0.65 1.50 0.66 0.66 0.55 0.51
Red 1.01 1.15 1.78 2.89 0.92 1.24 0.72 2.17
Bright Yellow 0.99 0.48 0.85 0.60 0.59 0.97 0.60 1.08
Green 0.66 0.33 0.53 0.22 0.42 0.68 0.64 0.86
Cyan 0.57 0.45 0.30 0.63 0.58 0.79 1.16 1.13
Deep Blue 1.70 0.98 0.35 0.90 1.48 1.57 0.80 0.90

Average
|difference| 0.78 0.62 0.60 0.89 0.62 0.83 0.57 0.86
3.5 Intercomparison results 39

Table 3.5: Partners’ colorimetric differences ∆E from NPL for the set of 16 color
tiles, specular excluded geometry. Fully corrected results, 10 degree observer, illuminant
D65 [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA

Glossy

Pale Grey 0.06 0.15 0.25 0.20 0.26 0.04 0.11 0.12
Mid Grey 0.16 0.20 0.26 0.33 0.31 0.11 0.18 0.06
Black 0.45 1.17 0.53 0.86 0.81 0.26 0.65 0.49
Red 0.44 1.15 0.90 0.73 1.07 0.77 0.86 2.10
Bright Yellow 0.66 0.48 0.47 0.99 0.72 0.72 0.69 1.09
Green 0.54 0.33 0.24 0.64 0.48 0.70 0.64 0.86
Cyan 0.56 0.45 0.27 0.73 0.57 0.93 1.07 1.15
Deep Blue 0.92 0.98 1.37 1.35 1.64 1.11 0.86 0.98

Average
|difference| 0.47 0.62 0.54 0.73 0.73 0.58 0.63 0.86
40 3. Measurement of surface color

Table 3.6: Partners’ colorimetric differences ∆E from the mean difference for the set
of 16 color tiles, specular included geometry. Uncorrected results, 10 degree observer,
illuminant D65 [43].

UJ LNE BAM CETO CSIC SP BCRA DELTA

Average
|difference| 0.27 0.32 1.09 0.31 0.22 0.27 0.25 0.20

% within 0.2
∆E of mean 56.3 43.8 25.0 37.5 75.0 62.5 43.8 62.5

% within 0.5
∆E of mean 75.0 81.3 43.8 75.0 93.8 81.3 93.8 93.8

% within 0.75
∆E of mean 87.5 87.5 56.3 93.8 93.8 87.5 100 100
3.5 Intercomparison results 41

Table 3.7: Partners’ colorimetric differences ∆E from the mean difference for the set
of 16 color tiles, specular included geometry. Fully corrected results, 10 degree observer,
illuminant D65 [43].

UJ LNE BAM CETO CSIC SP BCRA DELTA

Average
|difference| 0.22 0.20 0.78 0.21 0.22 0.26 0.18 0.23

% within 0.2
∆E of mean 43.8 50.0 31.3 62.5 75.0 50.0 56.3 50.0

% within 0.5
∆E of mean 93.8 100 62.5 81.3 93.8 87.5 100 100

% within 0.75
∆E of mean 100 100 62.5 100 93.8 100 100 100
42 3. Measurement of surface color

Table 3.8: Partners’ colorimetric differences ∆E from the mean difference for the set
of 16 color tiles, specular excluded geometry. Uncorrected results, 10 degree observer,
illuminant D65 [43].

UJ LNE BAM CETO CSIC SP BCRA DELTA

Average
|difference| 0.26 0.20 0.22 0.37 0.20 0.20 0.29 0.30

% within 0.2
∆E of mean 50.0 37.5 62.5 62.5 75.0 62.5 50.0 37.5

% within 0.5
∆E of mean 87.5 100 87.5 75.0 87.5 100 87.5 87.5

% within 0.75
∆E of mean 100 100 100 87.5 100 100 87.5 100
3.5 Intercomparison results 43

Table 3.9: Partners’ colorimetric differences ∆E from the mean difference for the set
of 16 color tiles, specular excluded geometry. Fully corrected results, 10 degree observer,
illuminant D65 [43].

UJ LNE BAM CETO CSIC SP BCRA DELTA

Average
|difference| 0.17 0.20 0.20 0.15 0.14 0.15 0.12 0.43

% within 0.2
∆E of mean 62.5 50.0 50.0 62.5 87.5 62.5 75.0 50.0

% within 0.5
∆E of mean 87.5 87.5 100 100 100 100 100 87.5

% within 0.75
∆E of mean 100 100 100 100 100 100 100 87.5
44 3. Measurement of surface color

3.6 Determination of colorimetric uncertainties

Generally, the result of a measurement is only a approximation or estimate of the
value of the specific quantity subject to measurement. Because of increasing demands
of measurement accreditation and quality systems there is a need to quote uncertainties
on all certified quantities. The result is complete only when accompanied a quantitative
statement of its uncertainty [8, 53].
The uncertainty of the result of a measurement generally consists of several compo-
nents. These components may be grouped into two categories according to the method
used to estimate their numerical values:

• Type A, evaluated by statistical methods.

• Type B, evaluated by other means.

The only Type A uncertainty in this intercomparison was the repeatability [43]. The
Type B uncertainties were:

• Uncertainty in the level of the absolute scales of diffuse reflectance and radiance
factor.

• Uncertainty in the spectral slope of the scales.

• Dark uncertainty.

• Linearity uncertainty.

• Wavelength scale uncertainty.

• Thermochromism uncertainty.

• Glossy to matt ratio uncertainty.

• Specular beam uncertainty.

• Gloss trap uncertainty.

The repeatability was determined by making several measurements. The standard

deviation was then used in the calculation of of the standard uncertainty.
Uncertainty in the level of the absolute scales of diffuse reflectance or radiance
factor comes from the traceability to the 100% levels. This is taken from a certificate
of calibration for a white standard. Uncertainty in the spectral slope of scales is
connected to absolute scale uncertainty. The scale may have a slope on it within the
uncertainty limits. This effect is assessed using a skew of half the scale uncertainty.
3.6 Determination of colorimetric uncertainties 45

The dark uncertainty is a combination of the electronic offset of the instrument and,
for diffuse reflectance, an optical offset due to a halo of scattered light surrounding the
sample beam and falling on the integrating sphere. Dark errors can be measured by
placing a gloss trap at the sample port of integrating sphere. Blocking the beam does
not give a true offset reading, because it does not quantify the effects of the halo of
stray light.
Photometric linearity uncertainty is connected to non-linearity of the detector. It
can be assessed from measurements of a white and a grey standards. Non-linearity
error was modelled using a polynomial and was set to zero at 0% and 100%.
Wavelength error cause changes in reflectance in regions of spectral slope, where
reflectance changes with wavelength. Wavelength error can be measured using a wave-
length standard or a known spectral line of a lamp. Thermochromism causes shift into
spectral data in a similar way as wavelength error.
Errors in the glossy to matt ratio are due to differences in efficiency of collection
of the integrating sphere with angle of reflectance. The error can be determined by
measuring glossy samples against matt samples with different integrating spheres.
The specularly reflected light may not be collected with the same efficiency as the
diffusely reflected light in the specular included geometry. Specular beam uncertainty
can be determined using a mirror and a calibrated matt white standard. In the specular
excluded geometry, the error due to incomplete absorption of the specular beam in the
gloss trap is known as the gloss trap error. This error can also be determined with a
mirror and a calibrated matt white standard.
Each uncertainty is characterised by assessed probability distribution for the un-
certainty. In a case of Gaussian (normal) distribution each component of uncertainty
was first divided by a coverage factor k, 1 or 2, which refers to confidence level of
approximately 65% or 95%, respectively. The total uncertainty was then combined in
quadrature from the components of uncertainty as

utotal = (u21 + u22 + . . . ), (3.21)

which was then multiplied by a coverage factor. In here coverage factor of k = 2 was
used.
The following components of uncertainties were determined for a UJ-data:

• repeatability

• photometric linearity uncertainty

• dark level uncertainty

46 3. Measurement of surface color

• gloss trap uncertainty

• specular beam uncertainty

Uncertainties in the level of the absolute scales and in the spectral slope of the
scales were left to calculate by co-ordinator since they provide partners absolute scales.
The wavelength uncertainty was not determined because of zero wavelength error in
measurement of the emission line of deuterium lamp. Also thermochromism effect was
neglected cause of agreed temperature limit ±1◦ C.
All uncertainty components were calculated using the determined errors. If the error
was corrected for in reflectance calculations, the uncertainty was that after correction.
If it was not corrected for, the uncertainty was the error itself.
In Tables 3.10–3.12 partners’ colorimetric uncertainties are shown for both specular
included and specular excluded geometries. For the UJ-data the most of the uncertainty
lies in a luminosity coordinate L∗ . This is due the fact that the photometric linearity
was the biggest error source. In Tables 3.13–3.14 partners’ colorimetric uncertainties
∗
combined to give ∆Eab for the intercomparison are shown.
3.6 Determination of colorimetric uncertainties 47

Table 3.10: Colorimetric uncertainties for partners tiles for a coverage factor of k = 2,
specular included geometry [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA NPL

Glossy

Pale L∗ 0.23 0.5 0.5 0.01 0.24 0.36 0.26 0.5 0.21
Grey a∗ 0.01 0.5 0.5 0.24 1.44 0.11 0.10 0.2 0.02
b∗ 0.00 0.5 0.5 0.07 0.75 0.34 0.19 0.2 0.02

Mid L∗ 0.35 0.5 0.5 0.00 0.18 0.40 0.29 0.5 0.16
Grey a∗ 0.01 0.5 0.5 0.14 1.10 0.08 0.18 0.2 0.02
b∗ 0.00 0.5 0.5 0.04 0.57 0.26 0.34 0.2 0.02

Black L∗ 0.28 0.5 0.5 0.27 0.12 0.32 0.81 0.5 0.16
a∗ 0.01 0.5 0.5 0.95 0.75 0.55 0.28 0.2 0.02
b∗ 0.00 0.5 0.5 0.98 0.38 0.17 0.79 0.2 0.02

Red L∗ 0.28 0.8 0.5 0.39 0.14 0.39 0.82 0.5 0.16
a∗ 0.02 0.8 0.5 1.45 1.04 0.25 0.38 0.2 0.15
b∗ 0.02 0.8 0.5 1.26 0.41 0.29 1.54 0.2 0.13

Bright L∗ 0.20 0.8 0.5 0.65 0.27 0.38 0.45 0.5 0.21
Yellow a∗ 0.14 0.8 0.5 2.23 1.70 0.32 0.47 0.2 0.11
b∗ 0.18 0.8 0.5 1.76 0.55 0.28 1.13 0.2 0.19

Green L∗ 0.34 0.5 0.5 0.47 0.17 0.39 0.23 0.5 0.15
a∗ 0.00 0.5 0.5 1.56 0.96 0.20 0.65 0.2 0.08
b∗ 0.01 0.5 0.5 1.58 0.46 0.28 1.13 0.2 0.13

Cyan L∗ 0.33 0.5 0.5 0.47 0.17 0.37 0.21 0.5 0.16
a∗ 0.03 0.5 0.5 1.60 0.95 0.32 0.73 0.2 0.13
b∗ 0.01 0.5 0.5 1.86 0.60 0.36 0.52 0.2 0.13

Deep L∗ 0.28 0.8 0.5 0.00 0.17 0.33 1.09 0.5 0.16
Blue a∗ 0.03 0.8 0.5 0.05 0.96 0.32 0.73 0.2 0.06
b∗ 0.05 0.8 0.5 0.02 0.46 0.26 0.64 0.2 0.11
48 3. Measurement of surface color

Table 3.11: Colorimetric uncertainties for partners tiles for a coverage factor of k = 2,
specular included geometry [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA NPL

Matt

Pale L∗ 0.23 0.5 0.5 0.01 0.26 0.43 0.26 0.5 0.21
Grey a∗ 0.01 0.5 0.5 0.24 1.57 0.11 0.10 0.2 0.05
b∗ 0.00 0.5 0.5 0.07 0.82 0.34 0.19 0.2 0.05

Mid L∗ 0.34 0.5 0.5 0.00 0.18 0.40 0.29 0.5 0.16
Grey a∗ 0.00 0.5 0.5 0.14 1.10 0.08 0.18 0.2 0.05
b∗ 0.00 0.5 0.5 0.04 0.57 0.26 0.34 0.2 0.05

Black L∗ 0.27 0.5 0.5 0.00 0.13 0.32 0.81 0.5 0.16
a∗ 0.00 0.5 0.5 0.01 0.77 0.55 0.28 0.2 0.05
b∗ 0.01 0.5 0.5 0.00 0.41 0.17 0.79 0.2 0.05

Red L∗ 0.30 0.8 0.5 0.00 0.15 0.39 0.82 0.5 0.16
a∗ 0.03 0.8 0.5 0.02 1.11 0.25 0.38 0.2 0.15
b∗ 0.03 0.8 0.5 0.01 0.46 0.29 1.54 0.2 0.11

Bright L∗ 0.20 0.8 0.5 0.00 0.29 0.38 0.45 0.5 0.21
Yellow a∗ 0.13 0.8 0.5 0.05 1.82 0.32 0.47 0.2 0.11
b∗ 0.19 0.8 0.5 0.02 0.60 0.28 1.13 0.2 0.19

Green L∗ 0.34 0.5 0.5 0.00 0.19 0.39 0.23 0.5 0.15
a∗ 0.01 0.5 0.5 0.02 1.07 0.20 0.65 0.2 0.08
b∗ 0.00 0.5 0.5 0.01 0.52 0.28 1.13 0.2 0.13

Cyan L∗ 0.33 0.5 0.5 0.00 0.18 0.37 0.21 0.5 0.16
a∗ 0.02 0.5 0.5 0.02 1.03 0.32 0.73 0.2 0.13
b∗ 0.02 0.5 0.5 0.01 0.65 0.36 0.52 0.2 0.13

Deep L∗ 0.28 0.8 0.5 0.00 0.13 0.33 1.09 0.5 0.16
Blue a∗ 0.02 0.8 0.5 0.01 0.78 0.32 0.73 0.2 0.06
b∗ 0.07 0.8 0.5 0.00 0.45 0.26 0.64 0.2 0.11
3.6 Determination of colorimetric uncertainties 49

Table 3.12: Colorimetric uncertainties for partners tiles for a coverage factor of k = 2,
specular excluded geometry [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA NPL

Glossy

Pale L∗ 0.25 0.5 0.5 0.00 0.23 0.40 0.26 0.5 0.20
Grey a∗ 0.00 0.5 0.5 0.04 1.38 0.10 0.14 0.2 0.02
b∗ 0.00 0.5 0.5 0.02 0.71 0.33 0.17 0.2 0.02

Mid L∗ 0.35 0.5 0.5 0.00 0.16 0.43 0.23 0.5 0.15
Grey a∗ 0.00 0.5 0.5 0.04 0.96 0.08 0.15 0.2 0.02
b∗ 0.00 0.5 0.5 0.01 0.50 0.25 0.22 0.2 0.02

Black L∗ 0.14 0.5 0.5 0.11 0.06 0.39 0.54 0.5 0.41
a∗ 0.04 0.5 0.5 0.36 0.35 0.04 0.26 0.2 0.02
b∗ 0.04 0.5 0.5 0.41 0.17 0.14 0.38 0.2 0.02

Red L∗ 0.22 0.8 0.5 0.13 0.16 0.39 0.74 0.5 0.18
a∗ 0.10 0.8 0.5 0.50 1.12 0.29 0.30 0.2 0.21
b∗ 0.03 0.8 0.5 0.45 0.53 0.44 1.59 0.2 0.47

Bright L∗ 0.21 0.8 0.5 0.00 0.26 0.39 0.37 0.5 0.20
Yellow a∗ 0.15 0.8 0.5 0.04 1.64 0.40 0.66 0.2 0.12
b∗ 0.07 0.8 0.5 0.01 0.56 0.60 0.39 0.2 0.27

Green L∗ 0.33 0.5 0.5 0.00 0.17 0.38 0.23 0.5 0.15
a∗ 0.05 0.5 0.5 0.02 0.93 0.25 0.43 0.2 0.10
b∗ 0.04 0.5 0.5 0.01 0.44 0.32 0.84 0.2 0.15

Cyan L∗ 0.32 0.5 0.5 0.00 0.16 0.37 0.37 0.5 0.16
a∗ 0.06 0.5 0.5 0.02 0.91 0.26 1.03 0.2 0.16
b∗ 0.02 0.5 0.5 0.01 0.56 0.33 0.63 0.2 0.14

Deep L∗ 0.19 0.8 0.5 0.00 0.18 0.32 0.61 0.5 0.39
Blue a∗ 0.05 0.8 0.5 0.02 1.02 0.38 1.01 0.2 0.48
b∗ 0.13 0.8 0.5 0.01 0.47 0.47 0.97 0.2 0.46
50 3. Measurement of surface color

Table 3.13: Partners’ colorimetric uncertainties combined to give a ∆E for the inter-
comparison for a coverage factor of k = 2, specular included geometry [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA

Glossy

Pale Grey 0.31 0.89 0.89 0.33 1.65 0.55 0.40 0.61
Mid Grey 0.39 0.88 0.88 0.22 1.26 0.51 0.51 0.60
Black 0.32 0.88 0.88 1.40 0.86 0.40 1.18 0.60
Red 0.38 1.41 0.90 1.98 1.16 0.60 1.80 0.63
Bright Yellow 0.43 1.42 0.92 2.93 1.83 0.65 1.34 0.65
Green 0.40 0.89 0.89 2.28 1.10 0.56 1.34 0.61
Cyan 0.41 0.90 0.90 2.51 1.16 0.65 0.95 0.62
Deep Blue 0.35 1.40 0.89 0.21 1.10 0.48 1.47 0.61

Matt

Pale Grey 0.32 0.89 0.89 0.21 1.58 0.57 0.40 0.61
Mid Grey 0.38 0.88 0.88 0.15 1.10 0.53 0.38 0.59
Black 0.44 0.96 0.96 0.69 0.57 0.58 0.82 0.71
Red 0.60 1.49 1.02 0.88 1.36 0.85 1.86 0.79
Bright Yellow 0.45 1.43 0.94 0.36 1.79 0.89 0.92 0.68
Green 0.41 0.90 0.90 0.24 1.07 0.60 1.00 0.62
Cyan 0.42 0.91 0.91 0.27 1.11 0.62 1.29 0.63
Deep Blue 0.81 1.59 1.16 0.77 1.37 1.03 1.71 0.96
3.7 Summary 51

Table 3.14: Partners’ colorimetric uncertainties combined to give a ∆E for the inter-
comparison for a coverage factor of k = 2, specular excluded geometry [43].

Tile UJ LNE BAM CETO CSIC SP BCRA DELTA

Glossy

Pale Grey 0.32 0.89 0.89 0.21 1.58 0.57 0.40 0.61
Mid Grey 0.38 0.88 0.88 0.15 1.10 0.53 0.38 0.59
Black 0.44 0.96 0.88 0.69 0.57 0.58 0.82 0.71
Red 0.60 1.49 1.02 0.88 1.36 0.85 1.86 0.79
Bright Yellow 0.45 1.43 0.94 0.36 1.79 0.89 0.92 0.68
Green 0.41 0.90 0.90 0.24 1.07 0.60 1.00 0.62
Cyan 0.42 0.91 0.91 0.27 1.11 0.62 1.29 0.63
Deep Blue 0.81 1.59 1.16 0.77 1.37 1.03 1.71 0.96

3.7 Summary
Although not all possible errors for the color measurements have been addressed, the
error correction method employed within the project has considerably improved the
agreement regarding measurements carried out by the partners for the 16 color tiles. At
the end of the project, over 92% of the measurements made by all the partners for all
∗
the tiles agreed to within 0.5 ∆Eab . This result fell slightly short of the target set at the
outset of the project but is considerably better than those agreements achieved in all
previous intercomparisons. In Tables 3.15–3.17 summary of percentage of colorimetric
differences are shown with errors within 0.2, 0.5 and 0.75 ∆E from the mean difference,
respectively.
52 3. Measurement of surface color

Table 3.15: Summary of percentage of Partners’ colorimetric differences ∆E from the

mean difference for the set of 16 color tiles, 10 degree observer, illuminant D65 . Total
percentage of errors within 0.2 ∆E from the mean difference [43].

Specular Included Specular Excluded

Correction applied geometry geometry

Uncorrected results (R1 ) 51% 55%

R1 + Wavelength correction (R2 ) 50% 58%
R2 + Sphere error correction (R3 ) 42% 63%
R3 + Linearity correction (R4 ) 52% 63%

Table 3.16: Summary of percentage of Partners’ colorimetric differences ∆E from the

mean difference for the set of 16 color tiles, 10 degree observer, illuminant D65 . Total
percentage of errors within 0.5 ∆E from the mean difference [43].

Specular Included Specular Excluded

Correction applied geometry geometry

Uncorrected results (R1 ) 80% 89%

R1 + Wavelength correction (R2 ) 79% 88%
R2 + Sphere error correction (R3 ) 91% 95%
R3 + Linearity correction (R4 ) 90% 95%
3.7 Summary 53

Table 3.17: Summary of percentage of Partners’ colorimetric differences ∆E from the

mean difference for the set of 16 color tiles, 10 degree observer, illuminant D65 . Total
percentage of errors within 0.75 ∆E from the mean difference [43].

Specular Included Specular Excluded

Correction applied geometry geometry

Uncorrected results (R1 ) 88% 97%

R1 + Wavelength correction (R2 ) 90% 97%
R2 + Sphere error correction (R3 ) 94% 98%
R3 + Linearity correction (R4 ) 95% 98%
 Chapter IV
Tristimulus integration

The permanent CIE Colorimetry Committee reviews from time to time the original
recommendations and makes those changes considered necessary. The work is done in
different working groups which are divided into technical committees. A working group
of USTC-1.3 of the CIE prepared, between 1981 and 1983, new recommendations for
the calculation of CIE tristimulus values. These recommendations state that the stan-
dard method of performing the integration forming the basic definition of tristimulus
values shall be by summation at a wavelength interval of 1 nm over the wavelength
range 360–830 nm, but that for most colorimetric purposes the approximation of sum-
mation at a 5nm interval over the range 380–780 nm should suffice. Recommendations
were also made concerning abridgement, interpolation, extrapolation, truncation and
the calculation of weighting factors, since measured data does not always fulfill the
requirements. Weighting factors are used in the calculation of color coordinates with
intervals which are not so dense, such as 10nm or 20 nm. Although tables of weighting
factors were not included in the recommendations to the CIE, they have since been
calculated in cooperation with the Working Group and published by the American
Society for Testing and Materials, ASTM [1].
CIE recommends the use of the method of calculation of weight sets for tristimulus
integration published by the ASTM in its Method E-308-85. In this method, the prepa-
ration of weight sets for any measurement interval or wavelength range by Lagrange
interpolation from standard tables of data at other intervals is described [2, 10, 11, 54].
In this chapter, we show that this recommendation is useless and that it should not be
used. Furthermore, its use may confuse the user and may lead to miscalculation.

54
4.1 The ASTM weighting method 55

4.1 The ASTM weighting method

The tristimulus values are defined by the equations


X=k β(λ)S(λ)x̄(λ)dλ

Y = k β(λ)S(λ)ȳ(λ)dλ (4.1)

Z = k β(λ)S(λ)z̄(λ)dλ.

The analytical functions of the three factors in the product are not known, so the
integrals should be evaluated alternatively. CIE recommends this be done by 1 nm
summation. Now the equations for such a summation are of the form

Y =k β(λ)S(λ)ȳ(λ)∆λ, (4.2)

where the values have the same meaning as in Eqs. 4.1. In order to solve Eq. 4.2 all
input values of the three functions should also be obtained at 1 nm intervals. Most
colorists will not wish to resort to measurement of the spectral radiance factors at 1
nm interval. In many cases, it would be sufficient to measure these functions at 10 nm
or 20 nm intervals. Those unmeasured 9 or 19 unit wavelengths radiance factors can be
interpolated by Lagrange interpolation, method recommended by the Working Group
VIII. The first and the last missing intervals should be interpolated quadratically using
three measured values and all interior values should be interpolated cubically using four
measured values. Lagrange interpolating coefficients may be calculated from

n
(r − ri )
Lj (r) = for j = 0, 1, . . . n, (4.3)
i=0
(rj − ri )
i=j

where n signifies the degree of the coefficients being calculated. The values of i and r
are the indices denoting the location on the abscissa of the tabular values among which
interpolation is to be carried out.
In the usual case where the measurement interval is uniform across the spectrum
56 4. Tristimulus integration

Eq. 4.3 simplifies to

(r − 1)(r − 2)(r − 3)
L0 = (4.4)
−6
r(r − 2)(r − 3)
L1 = (4.5)
2
r(r − 1)(r − 3)
L2 = (4.6)
−2
r(r − 1)(r − 2)
L3 = (4.7)
6

in the cubic case and to

(r − 1)(r − 2)
L0 = (4.8)
2
r(r − 2)
L1 = (4.9)
−1
r(r − 1)
L2 = (4.10)
2

in the quadratic case. In each of the above equations, indices r are substituted to
generate the necessary set of Lagrange interpolating coefficients. Suppose we have 20
nm data and we want to calculate Lagrange coefficients for those missing 19 values.
For intermediate intervals indices r now ranges from 1 + 1/20 to 2 − 1/20 at 1/20
steps. For last interval indices r ranges the same way, but for a first interval r takes
values from 0 + 1/20 to 1 − 1/20 again at 1/20 steps. Table 4.1 shows an example of
calculated Lagrange coefficients L0 , L1 , L2 , and L3 in a cubic interpolation for the case
of prediction of 19 values missing.
In Fig. 4.1 calculated Lagrange coefficients are shown both for the cubic and the
quadratic cases where data interval was 20 nm. The missing intermediate values may
be predicted from

n
P (r) = Li m i , (4.11)
i=0

where P is the value being interpolated, L is the Lagrange interpolation coefficient and
m is the measured value corresponding to the indices.
4.1 The ASTM weighting method 57

1.2

0.8 L0
coefficient value

0.6

0.4

L1
0.2

0 L2

−0.2
0 2 4 6 8 10 12 14 16 18 20
i

L1
0.8
coefficient value

0.6

0.4

0.2
L2

0
L3
L0
−0.2

0 2 4 6 8 10 12 14 16 18 20
i

Figure 4.1: Lagrange coefficients for the cubic case and the quadratic case where 19
missing values are interpolated.
58 4. Tristimulus integration

Table 4.1: Lagrange coefficients L0 , L1 , L2 , and L3 in a cubic case for 20 nm interval

data where 19 missing values are interpolated.

L0 L1 L2 L3

-0.0154 0.9726 0.0512 -0.0083

-0.0285 0.9405 0.1045 -0.0165
-0.0393 0.9042 0.1596 -0.0244
-0.0480 0.8640 0.2160 -0.0320
-0.0547 0.8203 0.2734 -0.0391
-0.0595 0.7735 0.3315 -0.0455
-0.0626 0.7239 0.3898 -0.0512
-0.0640 0.6720 0.4480 -0.0560
-0.0639 0.6181 0.5057 -0.0598
-0.0625 0.5625 0.5625 -0.0625
-0.0598 0.5057 0.6181 -0.0639
-0.0560 0.4480 0.6720 -0.0640
-0.0512 0.3898 0.7239 -0.0626
-0.0455 0.3315 0.7735 -0.0595
-0.0391 0.2734 0.8203 -0.0547
-0.0320 0.2160 0.8640 -0.0480
-0.0244 0.1596 0.9042 -0.0393
-0.0165 0.1045 0.9405 -0.0285
-0.0083 0.0512 0.9726 -0.0154

4.2 Spectral bandpass error

Spectral bandpass dependence occurs whenever the width of the passband is other than
infinitesimal [50]. The passband must be of sufficient width that the power passed to
the detector of the instrument is sufficient to lie within the dynamic range of the
detector. The most common situation in industrial equipment is that the bandpass
is both symmetrical and triangular in shape, and that the half-peak width is equal
to the measurement interval. This leads to a result where linear spectral radiance
within bandpass range does not cause any bandpass error. Instead, bandpass error
occurs if the spectral radiance is either concave or convex within the range of bandpass
function. This is because the instrument with a finite bandpass always measures the
4.2 Spectral bandpass error 59

spectral values too high in a concave situation compared to linear a extrapolant. In

a convex case, the situation is exactly opposite. An example of a triangular shape
bandpass and of a concave radiance function is shown in Fig. 4.2.

-9 -5 0 5 9

Figure 4.2: The representation of the triangular bandpass at one nanometer intervals
and its relationship to a concave radiance function. The dashed line represents the linear
extrapolant of the value of the function at the central wavelength.

Stearns has proposed that any wavelength index i can be estimated by

Rc (i) = −kRm (i − 1) + (1 + 2k)Rm (i) − kRm (i + 1), (4.12)

where Rc is the corrected radiance and the Rm is the measured radiance [50]. According
to Venable [54] the correct value for coefficient k is 0.083 in the case of a symmetrical
bandpass with a half-peak bandwidth equal to measurement interval. When correcting
either the first or last measured passband the correction is made by

Rc (i) = (1 + k)Rm (i) − kRm (i ± 1). (4.13)

The ASTM field test was done by Fairman [11]. In that research, 140 spectra were
used provided by cooperators. Wavelength range was from 380 nm to 780 nm with 10
60 4. Tristimulus integration

nm intervals. The spectra were considered to be arbitrarily corrected without regard

to their origin or geometry of measurement. In Table 4.2 summary of the results made
by Fairman are shown.

∗
Table 4.2: Summary of the results of all color difference ∆Eab calculations with and
without the ASTM weighting for 10 nm and 20 nm data [11].

Median 80-percentile

Interval None Weighting None Weighting

10 nm 0.145 0.004 0.253 0.011

20 nm 0.605 0.015 1.053 0.075

4.3 Materials and methods

Two data sets were used to analyze the ASTM weighting sets in the case of non
fluorescent color spectra. The first data set was measured with a Perkin Elmer Lambda
9 UV/VIS/NIR spectrophotometer with an external integrating sphere attached. This
data set consists of 1269 reflectance spectra of the matt Munsell chips. The second
data set was measured with a Perkin Elmer Lambda 18 UV/VIS spectrometer with
a Labsphere RSA-PE-18 reflectance spectroscopy accessory. In this set, there was a
total of 922 spectra of Pantone colors. Therefore the total number of spectra was 2191.
Both data sets were measured at 1 nm intervals from 380 nm to 780 nm. This data is
considered to be arbitrarily correct and free of bandpass error. Of course, this can be
done if we simply consider the original data to be correct, theoretically.
Let us first consider the effect of calculating CIELAB color coordinates with dif-
ferent wavelength intervals. In table 4.3, some statistics are shown for the calculated
∗
color differences ∆Eab with different wavelength intervals. Data intervals other than
1 nm was achieved by picking data with the appropriate interval from the original
data. The first value of the wavelength is 380 nm in every case. The last value is not
necessarily 780 nm if the new interval is not evenly divided within the original 400
nm wide wavelength band. Color differences are then compared with the original 1
nm data. Two different illuminants and observers were used which provided a total of
8764 differences for each interval.
4.3 Materials and methods 61

∗
Table 4.3: Color difference ∆Eab with different data intervals, each interval compared
to 1 nm reference data. There were a total of 8764 calculated color differences in each
interval.

Interval Mean Max Std Median 80-percentile

2 0.006 0.038 0.006 0.003 0.010

3 0.012 0.227 0.017 0.007 0.017
4 0.010 0.096 0.009 0.007 0.015
5 0.002 0.016 0.002 0.001 0.002
6 0.018 0.303 0.020 0.012 0.026
7 0.023 0.336 0.025 0.015 0.036
8 0.024 0.742 0.029 0.015 0.036
9 0.029 0.558 0.032 0.019 0.044
10 0.031 0.433 0.034 0.018 0.046
15 0.074 0.616 0.064 0.054 0.116
20 0.126 1.205 0.103 0.099 0.187
25 0.251 1.671 0.217 0.187 0.388
30 0.397 2.375 0.275 0.331 0.598
35 0.614 3.315 0.495 0.472 0.979
40 0.938 6.264 0.797 0.689 1.502
45 1.302 7.478 1.081 0.959 2.037
50 1.665 9.564 1.331 1.325 2.663

∗
One can see that the mean difference ∆Eab is still less than 0.5 when the wavelength
interval is over 30 nm. If all measurements should go under the limit of the human eye
∗
discrimination, 10 nm data would still be usable. The maximum ∆Eab error is 0.433
with a 10 nm interval. The last statistic 80 − percentile tells us under what limit 80%
of the data goes. Again the data interval can be almost 30 nm when 20% of the results
∗
are over the 0.5 ∆Eab limit value.
Let us now apply the ASTM weighting into the previous sets. Table 4.4 shows the
results with the ASTM weighting. If we now compare the results of Table 4.3 and Table
4.4 one can note only minimal changes. The weighting does not give significantly better
results. The situation is completely different in the results calculated by Fairman [11]
Table 4.2, where the effect of weighting was about tenfold. The reason may be in
the way in which the calculation was made. The starting point was the data of 140
62 4. Tristimulus integration

∗
Table 4.4: Color difference ∆Eab with different data intervals with ASTM weighting,
8764 differences in each interval.

Interval Mean Max Std Median 80-percentile

2 0.006 0.039 0.006 0.003 0.010

4 0.009 0.094 0.009 0.006 0.015
5 0.001 0.016 0.001 0.001 0.001
8 0.023 0.744 0.029 0.013 0.035
10 0.031 0.432 0.034 0.018 0.047
16 0.055 0.942 0.049 0.044 0.080
20 0.122 1.247 0.095 0.104 0.179
25 0.185 1.380 0.159 0.139 0.287
40 1.149 7.984 0.958 0.869 1.791
50 1.925 10.475 1.538 1.451 3.059

spectra with 10 nm intervals. From that data a new data set was formed by Lagrange
interpolation which was then set as the reference data to which the color difference
calculations were done.
Let us now simulate the calculations of the Fairman’s paper. First, 10 nm data is
acquired by picking every tenth reflectance value from the original 1 nm interval data
of the 2191 non fluorescent color spectra. Then, 1 nm reference data is interpolated
from the former 10 nm data. There is no need for any bandpass weighting if we assume
the original 1 nm data to be free of bandpass errors. Table 4.5 shows the results of the
statistics of the color differences with the ASTM weighted and unweighted data. Color
differences are calculated by comparing the 10 nm ASTM weighted and unweighted
data with the 1 nm data which was obtained by interpolation.
If we calculate the same color differences but use 1 nm original data as the reference
we obtain the results shown in Table 4.6. The difference between weighted and un-
weighted data is minimal. In fact, weighting has only a minor effect on the 10 nm data.
This is demonstrated in Table 4.7 where ∆E color difference is calculated between the
10 nm original and 10 nm the ASTM weighted data.
Furthermore, one can notice by comparing Tables 4.5 and 4.6 that the interpolated
1 nm data is much closer to the 10 nm data than to the original 1 nm data. Therefore
interpolation does not lead to the right original data. This can be realized by comparing
color differences of the original 1 nm data and the interpolated 1 nm data, as shown
4.3 Materials and methods 63

∗
Table 4.5: Color difference ∆Eab for the ASTM weighted and unweighted data, total
of 8764 differences. Color differences were calculated by comparing 10 nm data with 1
nm data which was obtained by interpolation.

Weighting Mean Max Std Median 80-percentile

None 0.0065 0.0793 0.0063 0.0046 0.0022

ASTM 0.0006 0.0028 0.0004 0.0005 0.0003

∗
Table 4.6: Color difference ∆Eab for the ASTM weighted and unweighted data, 8764
differences total. Color differences were calculated by comparing 10 nm data with the
original 1 nm data.

Weighting Mean Max Std Median 80-percentile

None 0.0309 0.4331 0.0342 0.0183 0.0271

ASTM 0.0310 0.4318 0.0342 0.0183 0.0278

∗
Table 4.7: Color difference ∆Eab for the original 10 nm data compared to the ASTM
weighted 10 nm data, 8764 differences in total.

Mean Max Std Median 80-percentile

0.0068 0.0805 0.0063 0.0049 0.0025

64 4. Tristimulus integration

in Table 4.8.

∗
Table 4.8: Color difference ∆Eab for the original 1 nm data compared to the 1 nm
interpolated data, 8764 differences in total.

Mean Max Std Median 80-percentile

0.0308 0.4319 0.0343 0.0180 0.0276

4.4 Results
It was shown that the use of the ASTM weighting functions in the case of non fluores-
cent reflectance spectra of matt color samples may not greatly affect the accuracy of
tristimulus integration with different wavelength intervals. However, the results in the
Fairman paper were shown to depend on the starting point of that research. It can not
be assumed that Lagrange interpolation can reliably accomplish 1 nm reference data
from the original 10 nm data in every case. Even with our instrument, which proved
to perform well in the previous chapter, it is not possible to interpolate correct 1 nm
data from correct 10 nm data. Hence, this becomes even more difficult with the 10 nm
interval instruments.
The error of absolute color coordinates of a given reflecting sample is mainly de-
pendent on the instrument that is used in the measurement. If accurate color values
are needed the performance of the instrument must be carefully studied at first, as
shown in the intercomparison study presented in the previous chapter. This makes the
ASTM standard E-308-85 useless, and we recommend it not be used. Furthermore, in
many quality control tasks where no exact absolute values are needed no weighting is
needed either.
 Chapter V
Thermochromism

The color of an object depends on the temperature of the object. This phenomena is
known as the thermochromic effect and it may be already noticed at room tempera-
ture if the temperature varies a few centigrades. Red and orange samples are especially
sensitive to temperature variation and may cause difficulties in precise color measure-
ments. This chapter shows, how the phenomenon is based on physical processes and
not only reflects the instability of red color pigments. Simple formulas are derived,
which explain the experimental data. We discuss the meaning of thermochromism
for color measurements, measure the magnitude of it and propose the experimental
conditions to avoid this effect [16, 19].

5.1 Background of the study

As pointed out in the previous chapters, accurate color measurements have become
more and more important during the past few decades. This is valid not only in
physical research, but also in industrial production, where the importance of accurate
measurements is mainly due to increased quality requirements set by the customers
of various goods. The development of technology enables more and more accurate
measuring systems. While the accuracy has improved one has noticed, that many un-
expected factors affect the color of an object. One of these factors is the temperature
of the sample. It is known that, for example, the reflectance of the ceramic reference
tiles used for calibration of colorimeters and spectrophotometers is temperature de-
pendent. This phenomenon is called thermochromism, which is a reversible change
of color of the sample as a function of temperature. Wyszecki and Stiles state some
general principles of thermochromism in the case of transmitting filters [60]. They
conclude that the spectral transmittance at a given wavelength that increases with in-
creasing wavelength usually decreases with increasing temperature. The more positive
the slope of the spectral transmittance curve, the greater the temperature effects as a
rule. Spectral transmittance that decreases with increasing wavelength is, in general, of

65
66 5. Thermochromism

minor importance, but if it becomes important it often causes a transmittance increase

with increasing temperature. Spectral transmittance curves varying only slightly with
wavelength are usually not sensitive to temperature variations. The data sheets for
SCHOTT colored glass filters give the shift of the half value of maximum transmit-
tance of the rising edge with increasing wavelength towards longer wavelengths in the
units of nm/K [52]. This value gradually increases with increasing wavelength from
0.07 nm at 400 nm to 0.17 nm/K at 700 nm. Malkin et al. have studied reflectance
spectra of ceramic color standards as a function of temperature [39]. They reported
that increasing the temperature of a sample caused a fall in reflectance which was
concentrated in the steeply rising or falling parts of the spectral reflectance. The part
of the spectral curve rising with increasing wavelengths was displaced towards longer
wavelengths, while the part falling with increasing wavelengths was displaced towards
shorter wavelengths on heating. The slope moves typically 0.1 nm/K. It is worth point-
ing out that the reflectance usually decreases as the temperature increases. The long
wavelengths reflecting e.g. yellow, orange and red samples are found to be affected
most, leading to the largest color differences in the CIELAB ∆E values. On the other
hand, neutral samples (grey, white, black) do not exhibit thermochrmism. The rea-
son is that the spectral reflectance curves of neutral colors have a small or zero slope.
Similar results for color differences of reflectance standards has been earlier reported
by Verrill et. al. [56] and Fairchild and Grum [9]. Instead of the wavelength shift
per nanometer or the color difference per centigrade, a third measure also exists: the
relative change of the optical density of the sample as a function of wavelength [55].
However, independent of the measure one likes to use, the main point is how much the
observed color depends on temperature.
The aim of this chapter is to derive a simple mathematical model for thermochromism,
and to investigate the magnitude of the effect. The experiments done for this investi-
gation as well as the published data discussed above confirm our model.

5.2 Absorbance, transmittance, and optical density

One knows from the standard literature that the color of an object under a given
illumination depends on the reflectance or transmittance spectrum of the sample. Re-
flectance and transmittance are the only quantities which, through absorbance depend
on temperature. In the following, we consider how the spectral properties may depend
on temperature, to see if the long wavelength end of the spectrum is affected more than
the short wavelength end of the spectrum. Luminescent materials might also exhibit
thermochromism, but they are not considered in this thesis. When light reaches a sur-
face of a nonfluorescent material, where neither nonlinear nor abnormal polarisation
5.2 Absorbance, transmittance, and optical density 67

effects appear, it holds that

1 = R(λ) + A(λ) + T (λ), (5.1)

where R(λ) is reflectance, A(λ) is absorption and T (λ) is transmittance of the sample.
In the following, we separately consider transmitting and opaque samples.

5.2.1 Transmitting samples

Let us consider external quantities. Light passing through an absorbing material at-
tenuates as

I(λ) = I0 (λ)e−µ(λ)d , (5.2)

where I0 (λ) is the incoming intensity on the front surface of the sample, I(λ) is the
intensity transmitted through the sample, d is the thickness of the sample and µ(λ) is
the absorption coefficient of the sample at wavelength λ. Definitely, µ(λ) is the main
quantity responsible for thermal effect.
We have the following relation between the transmittance and optical density

 
I(λ)
D = − log(T (λ)) = − log
I0 (λ)
µ(λ)d
= µ(λ)d log e = . (5.3)
ln 10

Hence, the changes in the absorption coefficient change the optical density, the ab-
sorbance and, thus, the transmittance of the sample.

5.2.2 Opaque samples

In the case of a glossy opaque sample, T (λ) is zero and the changes in absorbance di-
rectly change the reflectance. Now, the reflected part is comparable to the transmitted
part in transmitting samples and the optical density can be calculated from

 
IR (λ)
D = − log , (5.4)
I0 (λ)
68 5. Thermochromism

where IR (λ) is the reflected intensity.

On the other hand, the intensity of light falling upon some material attenuates
exponentially and, in the case of opaque materials, the light reflected from the material
experiences similar attenuation. Thus, Eq. (5.3) holds for the reflected light if d is the
distance the reflected light has passed in the material.
For translucent materials, the thickness of the material as well as the background
over which the sample is viewed affect the reflectance and the opacity of the material.
For all matt samples, the scattering properties of the material should be taken into
account through the Kubelka-Munk theory, although there is not a requirement for
a specimen to be matt for the Kubelka-Munk theory to hold. However, the simple
formulas given above are enough to investigate the thermochromism of the material.
We conclude that, as a good approximation, investigation of the thermal properties
of absorbance gives correct information on thermochromism for opaque, transmitting,
and translucent samples independent of whether the sample is glossy or matt.

5.3 Thermal effects

The depth from which the reflected light is coming is probably not affected by small
changes in temperature, but if µ(λ) changes as a function of temperature, it also affects
the reflected light distribution [14]. The absorption in the visible range is caused
by low level electron transitions, where an absorbed photon excites an electron from
one energy level to another with higher energy. The general case of an undisturbed
absorbing unit leads to two general solutions, which are of a Lorentzian and Gaussian
shape of µ(λ) and, hence, also of the absorption band. The Lorentzian band appears
at low temperatures and the Gaussian at high temperatures. The absorption peak
is homogeneously broadened if the absorbing units are indistinguishable. This means
that the absorption peak is symmetric in the energy scale, and in the frequency scale,
but not in the wavelength scale. Because the photon energy ε = hν where ν is the
frequency and h is the Planck’s constant, we will get a very simple relation between
energy and wavelength

1239.8
ε[eV] = . (5.5)
λ[nm]

Properties of the absorption band are usually given in electron volts. The Gaussian
shape of the absorption coefficient and hence the absorption band is [14] given by


4 ln 2 (ε − ε0 )2
µ (ε) = µmax exp − , (5.6)
(∆ε)2
5.3 Thermal effects 69

where ε0 is the energy of the absorption peak maximum and ∆ε is the full width at
half maximum absorption (called half width). In Fig. 5.1, an example of the Gaussian
peak is shown as a function of energy. Here ε0 =2.5 eV and ∆ε=0.4 eV. This means
that the left half width point (ε−1/2 ) is at energy 2.3 eV and the right half width
point (ε+1/2 ) corresponds to the energy 2.7 eV. In Fig. 5.2, the change of equi-energy

ε0
0.25

0.2
Absorbance

0.15

∆ε
0.1

0.05

0
ε−1/2 ε+1/2
1.5 2 2.5 3 3.5
Energy [eV]

Figure 5.1: Example of a Gaussian peak where ε=2.5 eV and ∆ε=0.4 eV. The positions
of the left and the right half width points are also shown.

absorbance peaks from the energy scale to the wavelength scale is demonstrated. If
thermal changes appear in the absorption band, they appear so that the integrated
absorption, which is the area of the absorption band, remains constant. In the case of
both line shapes this means that ∆εDmax remains constant. Here, Dmax is the peak
height. Gebhardt and Kuhnert first reported the change of the absorption bands as
a function of temperature [14]. The general behaviour of ∆ε at low temperatures
(near room temperature) is found to be proportional to T 1/2 , where T is the absolute
temperature in Kelvin. Thus, the temperature change from 20◦ C to 40◦ C (from 293
K to 313 K) causes that ∆ε of the absorption band becomes 1.034 times broader and
the peak height 1/1.034 = 0.968 times lower. In addition the peak maximum tends to
change towards smaller energies i.e, longer wavelengths as the temperature increases.
This increases the change of rising edge towards longer wavelengths and decreases the
changes of falling edge towards shorter wavelengths. The effect of this phenomenon
is usually small and the change of the position of absorption bands is not taken into
account in this study. The discussion above is valid only to a single absorption line.
Generally, a color spectrum in the energy-absorbance scale might be considered as
70 5. Thermochromism

0.9

0.8

0.7

Absorbance
0.6

0.5

0.4

0.3

0.2

0.1

0
1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2
Energy [eV]
1

0.9

0.8

0.7
Absorbance

0.6

0.5

0.4

0.3

0.2

0.1

0
400 450 500 550 600 650 700 750 800
Wavelength [nm]

Figure 5.2: Absorbance peaks changed from energy scale to wavelength scale.

a linear combination of Gaussian peaks given by Eq. (5.6). Figs. 5.3 and 5.4 give
examples of two different spectra, which are formed from a few Gaussian absorbance
curves.
These figures demonstrate how Gaussian absorption lines form absorbance spectra
in the energy scale, how they convert to the wavelength scale, and finally what the
resulting transmittance is in the wavelength scale. Note, that single absorption lines
are not symmetric in the wavelength scale. The rising and falling edges of a spectrum
5.3 Thermal effects 71

2.5

Absorbance
1.5

0.5

1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2

Energy [eV]

2.5

2
Absorbance

1.5

0.5

400 450 500 550 600 650 700 750 800

Wavelength [nm]

0.8
Transmittance

0.6

0.4

0.2

400 450 500 550 600 650 700 750 800

Wavelength [nm]

Figure 5.3: An example on the formation of the absorption and transmittance spectra
from three individual Gaussian Peaks.
72 5. Thermochromism

2.5

Absorbance
1.5

0.5

1.6 1.8 2 2.2 2.4 2.6 2.8 3 3.2

Energy [eV]

2.5

2
Absorbance

1.5

0.5

400 450 500 550 600 650 700 750 800

Wavelength [nm]

0.8
Transmittance

0.6

0.4

0.2

400 450 500 550 600 650 700 750 800

Wavelength [nm]

Figure 5.4: An example on the formation of the absorption and transmittance spectra
from three individual Gaussian Peaks.
5.3 Thermal effects 73

may be thought to consist of a single Gaussian peak. Thus, to understand the general
principles of thermochromism it is enough to study the properties of a single Gaussian
absorbance peak instead of a linear combination of them. The peaks forming the edges
have the major influence on the thermal properties of a spectrum whereas the other
peaks have a minor effect or none at all.
Let us consider in detail the properties of Gaussian absorbance curves. Absorbance
and optical density in the energy scale are

 
4 ln 2(ε − ε0 )2
D(ε) = Dmax exp − (5.7)
(∆ε)2

where Dmax = µmax d/ ln 10. In the visible range (1.59–3.3 eV) one can integrate the
absorption peak, and the result is area


π
A = Dmax ∆ε. (5.8)
4 ln 2
√
If the temperature changes, A remains constant and ∆ε = u T , where u is a constant
and T is given in Kelvins. Let ∆ε0 be the halfwidth of the peak at a temperature T0 .
Then at another temperature T the left and right halfwidth points ε±1/2 as a function
of temperature are


T
ε±1/2 (T ) = ε0 ± ∆ε0 . (5.9)
T0

Thus, the shift of the halfwidth points is

δε±1/2 (T ) = ε±1/2 (T ) − ε±1/2 (T0 )

 
1 T
= ± ∆ε0 −1
2 T0
1 ∆T
≈ ± ∆ε0 , (5.10)
4 T0

where T = T0 + ∆T . The approximation at the last line of Eq. (5.10) is quite valid up
to ∆T = 30 K.
74 5. Thermochromism

Let us next derive the shift formula in the wavelength scale, where the absorption
bands are not symmetric. Let λ±1/2 (T ) be the wavelengths corresponding to the points
ε±1/2 (T ). One can easily confirm that


1 1 1 ∆ε0 T
= ± , (5.11)
λ±1/2 (T ) λ0 2 hc T0

where λ0 is the wavelength corresponding to the peak maximum. The shift of the slope
is

δλ±1/2 (T ) = λ±1/2 (T ) + λ±1/2 (T0 )

 2
k λ±1/2 (T0 )
= , (5.12)
1 − kλ±1/2 (T0 )

where,
 
1 ∆ε0 T
k=± 1− . (5.13)
2 hc T0

Eqs. (5.12) and (5.13) describe exactly the slope shift in the wavelength scale. How-
ever, in many practical situations k is small and one gets from Eq. (5.12) as a first
approximation

1 ∆T ∆ε0  2
δλ±1/2 (T ) ∼
=∓ λ±1/2 (T0 ) . (5.14)
4 T0 hc

Let us finally calculate the shift of the transmittance curve by determining the shift
of the 50% transmittance points. For these points

1
= exp (− ln 10D) . (5.15)
2
√ √
Taking into account that T0 Dmax (T0 ) = T Dmax (T ) for two temperatures T0 and T
one can easily show that the half width point reduces to


1 T
ε±1/2 (T ) = ε0 ± ∆ε0 w (T ) , (5.16)
2 T0
5.4 Numerical results 75

where

1 1 T
w(T ) = √ ln Dmax + ln + 1.20055. (5.17)
ln 2 2 T0

Thus, one can directly use Eqs. (5.9)–(5.13) for transmittance (or reflectance) curves
by replacing ∆ε0 in these formulas by ∆ε0 w(T ).

5.4 Numerical results

We have calculated the change of reflectance and transmittance curves by generating
Gaussian absorption bands with different heights and half widths. We have also modi-
fied absorbance curves so that the integrated absorption remains constant so that half
width and maximum absorbance changes. The position of the peak maximum was
held constant. Modifications corresponded to temperature change from 300 K to 400
K. We used 0.2 eV and 0.4 eV as half widths, which are quite typical in solid mate-
rials. Maximum absorbance used was 2.0 and 4.0. This value effects the dynamics of
the transmitting filter. Maximum absorption 4.0 gives four orders of magnitude dy-
namics. Absorption curves were then converted to reflectance or transmittance curves
and the change of the rising or falling edge are read from half value of reflectance or
transmittance curves.
Fig. 5.5 shows the absolute shift of the halfwidth points of absorbance peaks as
a function of temperature change for three peak widths and two wavelengths. The
solid lines and the dashed lines correspond the shift of the rising and falling parts of
the peaks, respectively. The calculations were done using Eq. 5.12. One notes that
shift in nanometers is much higher for the edges in the red region than for a similar
peak in blue region. One also notes that there is a slight difference how the fallig and
rising parts of the peaks are moving. The reason is seen from Eq. 5.12. Note that
according to the approximative formula (Eq. 5.14) positive and negative slope exhibit
an equal shift. Figure 5.6 shows the absolute shift of the absorbance half width points
as a function of wavelength. The reference temperature is 300 K, and the peak width
at this temperature was set to 0.4 eV. The curves a–e correspond the behavior of the
material when the temperature change is 50–10 K. The solid lines and the dashed lines
correspond the shift of the rising and falling parts of the peaks, respectively. Fig. 5.6
shows clearly, that blue samples are not as sensitive as the red samples for temper-
ature changes. Again, exact formulas were used to show the difference between the
positive and negative slopes. On the other hand, the approximative formulas, which
do not show any difference between the negative and positive slopes, are valid only on a
limited temperature range. Fig. 5.7 shows the shift of the halfwidth points in nanome-
ters/K. The curves were calculated for a temperature change 10 K when the reference
76 5. Thermochromism

temperature is 300 K. The effect of the peak height and width is demonstrated, too.
As can be seen, by increasing the half widths of the absorption band or height of the
absorption peak, the changes in rising and falling edges become bigger.
10

Shift of the halfwidth points [nm]
9
T0 = 300 K
a) λ = 600 nm, ∆ε = 0.4 eV
8 b) λ = 600 nm, ∆ε = 0.2 eV a
c) λ = 400 nm, ∆ε = 0.4 eV
7
d) λ = 600 nm, ∆ε = 0.1 eV
6 e) λ = 400 nm, ∆ε = 0.2 eV
f) λ = 400 nm, ∆ε = 0.1 eV
5

4 b

3 c

2 d
e
1
f
0
0 10 20 30 40 50 60 70 80 90 100
Temperature change [K]

Figure 5.5: Shift of the halfwidth points as a function of temperature. The solid
and dashed curves correspond to the shift of the rising and falling parts of the spectra,
respectively.

5.5 Thermochromic measurements

Red colored Pantone
paper samples, red, orange, and yellow Labsphere
secondary
standards and cyan, green, yellow and red ceramic tiles were the samples used in this
study. These samples have a common spectral feature: they reflect less or about 10
percent for short wavelengths before the steeply rising part of the spectrum. For long
wavelengths these samples reflect well over 70 percent. The reflectance measurements
at different temperatures were made on a Perkin Elmer Lambda 18 spectrophotometer
equipped with a 150 mm integrating sphere. All of the measurements were made in
the 0/d geometry and in the specular included mode. The scanning speed was 480
nm/min, bandwidth 2 nm, and the measuring range 380–780 nm.
Before measuring the thermochromic samples, the errors of the spectrophotometer
were carefully investigated and the instrument were calibrated using the NPL stan-
dards. Sample imperfections (size and position of the illuminating beam on the sam-
ple) were investigated but found to be negligible in this study. The purpose of this
calibration procedure was to guarantee that the spectrophotometer is reliable in the
5.5 Thermochromic measurements 77

8 T0 = 300 K, ∆ε = 0.4 eV

Shift of the halfwidth points [nm]

7 a) ∆T = 50 K a
b) ∆T = 40 K
6
c) ∆T = 30 K
b
5
d) ∆T = 20 K
e) ∆T = 10 K
4 c

3
d

2
e
1

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

Figure 5.6: Shift of the halfwidth points as a function of wavelength. The solid and
dashed curves correspond to the shift of the rising and falling parts of the spectra,
respectively.

0.35

T0 = 300 K, ∆T = 10 K
0.3
Shift of the halfwidth points [nm/K]

a) ∆ε = 0.4 eV, Dmax = 4 eV

a
0.25 b) ∆ε = 0.4 eV, Dmax = 2 eV
c) ∆ε = 0.2 eV, Dmax = 4 eV
0.2 d) ∆ε = 0.2 eV, Dmax = 2 eV b

0.15
c

0.1 d

0.05

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

Figure 5.7: Shift of the halfwidth points in nanometers/Kelvins as a function of wave-

length. The solid and dashed curves correspond to the shift of the rising and falling parts
of the spectra, respectively.
78 5. Thermochromism

measurements of the thermochromic effect. The measurements at different temper-

atures were made as follows. Pantone
samples were heated with a brass cylinder
using an electric resistor. Labsphere SCS standards were heated with another heating
plate and a thermo resistor system was used with the ceramic tiles. The samples were
heated up to 94 centigrades outside the photometer to avoid heating of the instrument.
The temperature of the samples was measured with a calibrated thermopile during the
spectral measurements. The average measurment accuracy of the surface temperature
of the samples was less than one degree in all measurements. The repeatability of our
measurements became excellent after these calibration and temperature checking pro-
cedures. Spectral and colorimetric changes caused by the thermochromic effect were
noticed if the temperature was changed by about two degrees. The corrected spectra
were used to calculate the CIELAB coordinates and the color differences by standard
means. We need to point out, that although it is well known that CIELAB is not
linear it is generally accepted for the expression of color differences. In this coordinate
system, an average human eye is able to distinguish ∆E differences higher than 0.5
units, as pointed out in the beginning of this thesis.

5.6 Experimental results

Table 5.1 shows the results for a typical Pantone
sample (red 032 U). One can note
that all of the CIELAB coordinates decrease while the temperature increases. The
color of this typical paper sample changes 0.093 ∆E units per centigrade. If we define
the wavelength shift observing the reflectance value 50 percent at the steeply growing
part of the spectrum we found that that this part linearly moves towards the longer
wavelengths 0.1 nm/deg, which is of the same order as that for the ceramic tiles.
Tables 5.2 and 5.3 shows the results for two typical Labsphere secondary standards.
One can see decreasing luminance coordinate values with an increase in temperature.
Note, that especially in case of the red sample, the color difference compared to the
reference temperature increases very rapidly, and in typical living surroundings the
color change of this kind of samples can be noticed by any person with normal color
vision. The average ∆E/deg change of the red and orange samples was 0.107 and 0.088
∆E/deg, respectively. The color change of our yellow samples was clearly smaller.
The wavelength change of the red, orange, and yellow sample was 0.11, 0.09, and 0.06
nm/deg, respectively. Both the wavelength change and the color difference change are
linear functions of temperature on the temperature range measured.
In Table 5.4 the results of the thermochromic measurements of the ceramic tiles of
cyan, green, yellow and red are shown. Calculated changes for the wavelength shift for
the rising edge were approximately 0.05, 0.12, 0.18 and 0.15 nm/K, respectively. Fig.
5.8 shows the measured reflectance spectra for the red ceramic tile at temperatures
27◦ C–80◦ C.
5.7 Conclusions 79

Table 5.1: CIELAB values for the Pantone Red 032 U sample at four temperatures
[◦ C]. Color difference ∆E was calculated from the lowest temperature curve.

Temperature
[◦ C] L∗ a∗ b∗ ∆E

28 55.75 58.42 27.63

35 55.46 58.17 27.26 0.53
40 55.20 57.98 26.88 1.03
45 55.00 57.78 26.55 1.46

Table 5.2: CIELAB values for the red Labsphere secondary standard SCS-R-010 sample
at six different temperatures [◦ C]. Color difference ∆E was calculated from the lowest
temperature curve.

Temperature
[◦ C] L∗ a∗ b∗ ∆E

26 47.31 50.38 25.53

30 47.40 49.76 25.04 0.80
35 47.04 49.51 24.75 1.20
40 46.79 49.27 24.43 1.65
45 46.65 48.92 24.11 2.14
50 46.46 48.50 23.68 2.77

5.7 Conclusions
The magnitude of the thermochromic effects of transmitting and reflecting materials
has been calculated. If only the change of half width of the absorption curve as a
function of temperature is considered, we will obtain the result that the rising edge with
increasing wavelength shifts towards longer wavelengths and the falling edge towards
shorter wavelengths. The greater the slope (high absorbance) is, the bigger the change
80 5. Thermochromism

Table 5.3: CIELAB values for the orange Labsphere secondary standard SCS-O-010
sample at six different temperatures [◦ C]. Color difference ∆E was calculated from the
lowest temperature curve.

Temperature
[◦ C] L∗ a∗ b∗ ∆E

26 65.28 53.60 52.68

30 65.10 53.74 52.55 0.26
35 64.83 53.89 52.20 0.72
40 64.65 53.98 51.63 1.27
45 64.45 54.06 51.28 1.69
50 64.29 54.17 51.02 2.02

Table 5.4: CIELAB ∆E color difference values for the ceramic tiles at six different
temperatures [◦ C]. Color difference ∆E was calculated from the lowest temperature.

Cyan Green Yellow Red

T [◦ C] ∆E T [◦ C] ∆E T [◦ C] ∆E T [◦ C] ∆E

27 28 28 27
38 0.45 40 0.95 40 1.73 39 1.26
48 0.83 50 1.47 50 2.95 52 3.14
62 1.39 60 2.17 60 4.03 60 3.90
72 1.67 70 2.77 70 5.08 68 4.80
82 1.99 83 4.24 80 6.41 80 6.17

is. If we take into account the shift of the peak position this enhances the shift of
the rising edge and diminishes the change of falling edge. In some cases, the effect of
peak a position change towards longer wavelengths can be greater than a shift in the
5.7 Conclusions 81

90 50

80 49

48
70

47
60
Reflectance [%]

Reflectance [%]
46
50
45
40
44

30
43

20
42

10 41

0 40
400 450 500 550 600 650 700 750 600 605 610 615 620 625 630 635 640
Wavelength [nm] Wavelength [nm]

Figure 5.8: Curves from left to right represent the reflectance spectra of a red ceramic
tile measured at temperatures of 27, 39, 52, 60, 68 and 80◦ C, respectively.

falling edge towards longer wavelengths. This can happen if the slope of the falling
edge is small (low absorption) and total shift can be even towards long wavelengths as
proposed by Wyszecki and Stiles [60]. In the case of a rising edge the shift is always
towards longer wavelengths. The changes become bigger as the wavelength increases.
This means the biggest color differences occur for yellow, orange and red samples as
found in other studies.
In conclusion the thermal shift of a transmittance or reflectance curve edge is di-
rectly proportional to the temperature change and the slope of the spectrum as well as
directly proportional to the square of the wavelength. Because of thermochromism, new
recommendations for color measurements require the temperature during measurement
to be set to a predefined temperature e.g. (25 ± 1)◦ C.
 Chapter VI
Conclusions

Surface color measurement is of importance in a very wide range of industrial appli-

cations including paint, paper, printing, photography, textiles and plastics. For de-
manding color measurement the spectral approach is often needed. The most accurate
measurements of non fluorescent surface color are made with a spectrophotometer. The
method used internationally for specifying color has been recommended by the CIE.
These recommendations were shortly described and presented in this thesis.
The human eye can perceive color difference as small as 0.5 CIELAB units and thus
distinguish millions of colors. This 0.5 unit difference should be a goal for precise color
measurements. By the CIE recommendations the only geometry for reflectance mea-
surement is 0/d geometry, other geometries give only values of reflectance factors. The
use of a 0/d geometry will require the use of an integrating sphere to collect reflected
flux. In practical color measurement situations, an integrating sphere is never ideal
but hemispherical. Thus, absolute reflectance can not be collected either, ideally. In
this thesis, methodologies for measuring and correcting errors in color measurements
are discussed. These results are part of the European surface color measurement har-
monisation project. It was shown that it is possible to improve upon the agreement of
color measurement results where 16 samples of the same colors were measured in nine
laboratories of EU countries.
The sampling interval in a tristimulus integration affects the accuracy of a color
coordinate performance. In a publication of the American Society for Testing and
Materials, ASTM, in its method E-308-85 a method for the preparation of weight
sets for any measurement interval or wavelength range by Lagrange interpolation from
standard tables of data at other intervals is described. This method is recommended
by the CIE, although it is not included in CIE recommendations. In this thesis, this
ASTM method is tested with a large set of non fluorescent reflectance colors from
the Munsell and Pantone basis. The usefulness of this method was found to be quite
minimal and we recommended not to use it.

82
 83

The color of an object depends on the temperature of the object. This phenomena
is known as the thermochromic effect and it may be noticed already at room tem-
perature if the temperature varies by a few centigrade. Red and orange samples are
especially sensitive to temperature variation and may cause difficulties in precise color
measurements. In this thesis, it has been shown how this phenomenon is based on
physical processes and, that it does not only reflect the instability of color pigments.
Simple formulas are derived, which explain the experimental data. The meaning of
thermochromism for color measurements is also discussed, the magnitude of it is mea-
sured and experimental conditions to avoid this effect is proposed to be ±1◦ C.
 References

[1] ASTM Standards on Color and Appearance measurement, Standard Test Method
for Computing the Colors of Objects by Using the CIE System, 4th ed., 233–259,
(ASTM, Philadelphia, 1994).
[2] J. F. W. Billmeyer and H. S. Fairman, “CIE Method for Calculating Tristimulus
Values,” Color Research and Application 12, 27–36 (1987).
[3] R. M. Boynton, Human Color Vision (Holt, Rinehart & Winston, New York,
1979).
[4] K. M. Braun and M. D. Fairchild, “Testing Five Color-Appearance Models for
Changes in Viewing Conditions,” Color Research and Application 22, 165–173
(1997).
[5] CIE Publication 15.2, Colorimetry, 2nd Edition, Central Bureau of the CIE, (Vi-
enna, 1986).
[6] CIE Publication No 142-2001, CIE Technical Report: Improvement to Industrial
Colour-Difference Evaluation, Central Bureau of the CIE, (Vienna, 2001).
[7] F. J. J. Clarke and J. A. Compton, “Correction methods for integrating-sphere
measurement of hemispherical reflectance,” Color Research and Application 11,
253–262 (1986).
[8] P. J. Clarke, A. R. Hanson, and J. F. Verrill, “Determination of colorimetric un-
cerntainties in the spectrophotometric measurement of colour,” Analytica Chimica
Acta 380, 363–367 (1999).
[9] M. D. Fairchild and M. Grum, “Thermochromism of ceramic reflectance tiles,”
Applied Optics 24, 3432–3433 (1985).

84
References 85

[10] H. S. Fairman, “The Calculation of Weight Factors for Tristimulus Integration,”

Color Research and Application 10, 199–203 (1985).

[11] H. S. Fairman, “Results of the ASTM Field Test of Tristimulus Weighting Func-
tions,” Color Research and Application 20, 44–49 (1995).

[12] H. S. Fairman and H. Hemmendinger, “Stability of Ceramic Color Reflectance

Standards,” Color Research and Application 23, 408–415 (1998).

[13] J. Guild, “The colorimetric properties of the spectrum,” Philosophical Transac-

tions of the Royal Society A 230, 149–187 (1931).

[14] W. Hayes and A. M. Stoneham, Defects and Defect Processes in Nonmetallic

Solids (Wiley & Sons, New York, 1991).

[15] J. Hiltunen, T. Jaaskelainen, and J. P. Parkkinen, “Accurate spectral color mea-

surements,” Intelligent Robots and Computer Vision XVIII: Algorithms, Tech-
niques, and Active Vision, Proc. SPIE 3837, 202–209 (1999).

[16] J. Hiltunen, J. Mutanen, T. Jaaskelainen, and J. Parkkinen, “Thermochromism

in color measurement,” in Proceedings of the 9th Congress of the International
Colour Association (Rochester, 2001), in press.

[17] J. Hiltunen, R. Silvennoinen, T. Jaaskelainen, K. Nygren, and J. Parkkinen, “Re-

flectance spectra of pine grown in polluted environment,” in Proceedings of the
XXVIII annual conference of the Finnish physical society, M. Hyvonen-Dabek,
ed. (Helsinki, 1994).

[18] R. W. G. Hunt, Measuring Colour (Ellis Horwood Ltd., Chichester, 1987).

[19] J. Hiltunen, P. Silfsten, T. Jaaskelainen, and J. P. S. Parkkinen, “A Qualita-

tive Description of Thermochromism in Color Measurement”, Color Research and
Application, in press.

[20] T. Jaaskelainen, J. Hiltunen, H. Laamanen, and J. Parkkinen, “Accurate Spec-

tral Measurement and Reconstruction,” in Proceedings of the Second Interna-
tional Symposium on Multispectral Imaging and High Accurate Color reproduction
(Chiba, 2000).

[21] T. Jaaskelainen, J. Hiltunen, R. Silvennoinen, K. Nygren, and J. Parkkinen, “Spec-

tral forest monitoring on VIS-NIR region,” in International Symposium on Meth-
ods and Equipment for Environmental Monitoring (St. Petersburg, 1995).
86 References

[22] T. Jaaskelainen, J. Parkkinen, J. Hiltunen, and M. Hauta-Kasari, “Spectral ap-

proach to color and color images,” in Proceedings of the ICO Topical Meeting on
Optics for Sustainable Development (Dakar, 2000).

[23] T. Jaaskelainen, R. Silvennoinen, J. Hiltunen, and J. P. S. Parkkinen, “Classifica-

tion of reflectance spectra of pine, spruce and birch,” Applied Optics 33, 2356–2362
(1994).

[24] D. B. Judd, “Reduction of data on mixture on color stimuli,” Bureau of Standards

Journal of Research 4, 515–548 (1930).

[25] S. J. Kishner, “Effect of spectrophotometric errors on color difference,” Journal

of the Optical Society of America 67, 772–779 (1977).

[26] R. G. Kuehni, “Hue Uniformity and the CIELAB Space and Color Difference
Formula,” Color Research and Application 23, 314–322 (1998).

[27] R. G. Kuehni, “Analysis of Five Sets of Color Difference Data,” Color Research
and Application 26, 141–150 (2001).

[28] R. G. Kuehni, “Color Space and Its Divisions,” Color Research and Application
26, 209–222 (2001).

[29] R. G. Kuehni, “From Color-Matching Error to Large Color Differences,” Color

Research and Application 26, 384–393 (2001).

[30] R. Lenz, M. Österberg, J. Hiltunen, T. Jaaskelainen, and J. Parkkinen, “Exploring

color spaces with unsupervised filters and gaussian quadrature,” in Proc. SSAB
Conference (Linkoping, 1995).

[31] R. Lenz, M. Österberg, J. Hiltunen, T. Jaaskelainen, and J. Parkkinen, “Unsuper-

vised filtering of color spectra,” in Proceedings of the 9th Scandinavian Conference
of Image Analysis (Uppsala, 1995).

[32] R. Lenz, M. Österberg, J. Hiltunen, T. Jaaskelainen, and J. Parkkinen, “Unsu-

pervised filtering of color spectra,” Journal of the Optical Society of America A
13, 1315–1324 (1996).

[33] C. J. Li, M. R. Luo, and R. W. G. Hunt, “A Revision of the CIECAM97s,” Color

Research and Application 25, 260–266 (2000).

[34] M. R. Luo, G. Cui, and B. Rigg, “The Development of the CIE 2000 Colour-
Difference Formula: CIEDE2000,” Color Research and Application 26, 340–350
(2001).
References 87

[35] M. R. Luo and R. W. G. Hunt, “The Structure of the CIE 1997 Colour Appearance
Model (CIECAM97s),” Color Research and Application 23, 138–146 (1998).
[36] M. R. Luo, M.-C. Lo, and W.-G. Kuo, “The LLAB(l:c) Colour Model,” Color
Research and Application 21, 412–429 (1996).
[37] J. Luostarinen and J. Hiltunen, “On-line spectral measurement for grading of
beech parquet blocks,” in Proceedings of the Finnish optics days 1998 (Oulu, 1998).
[38] J. Luostarinen, J. Hiltunen, and K. Paukkonen, “Use of colour data in grading of
beech parquet blocks,” in 3rd International Wood Scanning Seminar (Skellefteå,
1998).
[39] F. Malkin, J. A. Larkin, J. F. Verrill, and R. H. Wardman, “The BCRA-NPL
Ceramic Standards, Series II - Master spectral reflectance and thermochromism
data,” Journal of the Society of Dyers and Colourists 113, 84–94 (1997).
[40] H. Minato, M. Nanjo, and Y. Nayatani, “Colorimetry and its Accuracy in the Mea-
surement of Fluorescent Materials by the Two-Monochromator Method,” Color
Research and Application 10, 84–91 (1985).
[41] Y. Nayatani, “Examination of Adaptation Coefficients for Incomplete Chromatic
Adaptation,” Color Research and Application 22, 156–164 (1997).
[42] Y. Nayatani, H. Sobagaki, K. Hashimoto, and T. Yano, “Field Trials of a Nonlinear
Color-Appearance Model,” Color Research and Application 22, 240–258 (1997).
[43] NPL report COEMS33, Harmonisation of National Scales of Surface Colour Mea-
surements, (Luxembourg, 2000).
[44] Project report EUR 19552EN, Good Practice Guide to Surface Colour Measure-
ments, (Luxembourg, 2000).
[45] Report EUR 14982EN, Intercomparison of colour measurements, Synthesis report,
(Luxembourg, 1993).
[46] R. Silvennoinen, T. Jaaskelainen, K. Nygren, J. Hiltunen, and J. Parkkinen, “Clas-
sification of Reflectance Spectra of Pine, Spruce and Birch,” in First Finnish Con-
ference of Environmental Sciences, J. Tuomisto and J. Ruuskanen, eds. (Kuopio,
1993).
[47] R. Silvennoinen, T. Jaaskelainen, K. Nygren, J. Hiltunen, and J. Parkkinen, “Re-
flectance spectra of pine, spruce and birch grown in polluted environment,” in
Optical Methods in Biomedical and Environmental Sciences, H. Ohzu and S. Ko-
matsu, eds. (Elsevier Science B. V., Tokyo, 1994).
88 References

[48] R. Silvennoinen, T. Jaaskelainen, K. Nygren, J. Hiltunen, and J. P. S. Parkkinen,

“Temporal, spatial and environmental classification of pine reflectance spectra,”
Environmental Science and Technology 29, 1456–1459 (1995).

[49] N. I. Speranskaya, “Determination of spectrum colour coordinates for twenty-

seven normal observers,” Optics and Spectroscopy 7, 424–428 (1959).

[50] E. I. Stearns and R. E. Stearns, “An example of a method for correcting data for
bandpass error,” Color Research and Application 13, 257–259 (1988).

[51] W. S. Stiles and J. M. Burch, “N. P. L. Colour-matching investigation: Final

report,” Optica Acta 6, 1–26 (1959).

[52] URL: http://www.besoptics.com/html/schott color filters.html (valid

15th October 2001).

[53] URL: http://www.ukas.com/pdfs/LAB12.PDF (valid 11th December 2001).

[54] W. H. Venable, “Accurate tristimulus values from spectral data,” Color Research
and Application 14, 260–267 (1989).

[55] J. Verrill, “Advances in spectrophotometric transfer standards at NPL,” in Spec-

trophotometry, Luminescence and Colour; Science and Compliance, C. Burgess
and D. G. Jones, eds. (Amsterdam, 1995).

[56] J. F. Verrill, J. A. Compton, and F. Malkin, “Thermochromism of ceramic color

standards,” Applied Optics 25, 3011 (1986).

[57] N. J. Wade, A Natural History of Vision, 2nd ed. (The MIT Press, Cambridge,
1999).

[58] B. A. Wandell, “Measurement of Small Color Diffrences,” Psychological Review

89, 281–302 (1982).

[59] W. D. Wright, “A re-determination of the trichromatic coefficients of the spectral

colours,” Transactions of the Optical Society 30, 141–164 (1929).

[60] G. Wyszecki and W. S. Stiles, Color Science: Concepts and Methods, Quantitative
Data and Formulae, 2nd ed. (John Wiley & Sons, New York, 1982).