Computer Graphics Unit

Manchester Computing Centre

University of Manchester

Department of Computer Science

University of Manchester

Colour
in
Computer Graphics

Student Notes

C. Lilley
F. Lin
W.T. Hewitt
T.L.J.Howard

ITTI Computer Graphics and Visualisation

 Table of Contents

1 Introduction ............................................................................. 1

1.1 Contents and scope ............................................................ 1

2 Seeing in colour ...................................................................... 3

2.1 The electromagnetic spectrum .......................................... 3

2.2 Spectra ................................................................................ 4

2.3 The eye ................................................................................ 5

2.4 The retina ........................................................................... 7

2.5 Receptor cells ..................................................................... 8

2.6 Colour reception ................................................................. 9

2.7 The fovea .......................................................................... 12

3 Measuring colour ................................................................. 13

3.1 Colour matching ............................................................... 13

3.2 A standardised observer .................................................. 14

3.3 Coloured objects ............................................................... 16

3.4 Emissive colour ................................................................ 16

3.5 Illuminants ....................................................................... 18

3.6 Reflective colour ............................................................... 21

3.7 Chromaticity .................................................................... 22

3.8 Atypical colour response .................................................. 26

4 Colour models ........................................................................ 31

4.1 Why use colour models? ................................................... 31

4.2 Primary colours ................................................................ 31

4.3 CIE colour models ............................................................ 34

4.4 Device dependent models ................................................ 38

University of Manchester i
 4.5 Other colour models ......................................................... 42

5 Colour output ........................................................................ 49

5.1 Displaying colour ............................................................. 49

5.2 Colour video ...................................................................... 55

5.3 Broadcasting colour ......................................................... 61

5.4 Coping with insufficient colours ...................................... 63

5.5 Printing in colour ............................................................. 69

5.6 Colour photography ......................................................... 73

5.7 Gamut mapping ............................................................... 74

6 Usage of colour ...................................................................... 79

6.1 When to use colour ........................................................... 79

6.2 Selecting a colour model .................................................. 79

6.3 Colour schemes ................................................................ 80

6.4 Interpolation .................................................................... 82

A Gamma correction ............................................................ 83

A.1 Determining gamma ....................................................... 83

A.2 Direct measurement ........................................................ 83

A.3 Visual calibration ............................................................ 84

B Monitor calibration .......................................................... 87

C Glossary ............................................................................... 89

ii Computer Graphics and Visualisation

1 Introduction

A good understanding of colour is essential for effective use of computer graphics.

This module describes the science of colour as it applies to computer graphics
and visualisation.

1.1 Contents and scope

This module is divided into five sections.

Firstly, we look at how colour is seen. This draws together information from such
diverse disciplines as physics, optics, physiology, neurology and psychology to
show that colour is an internal, subjective sensation rather than an external, ob-
jective entity. This helps explain just what colour is.

Given the biological basis of colour, how can it be measured and standardised?
The second section explains how colour is measured and introduces the CIE in-
ternational standard, used to define colour. This provides the vital link between
biological sensation and physical measurement. Examples are also given of how
colour measurements can be used and manipulated, such as predicting the result
of a colour mixture or designing displays for people with defective colour vision.

An abstraction called a colour model is used to specify colour. The third section
explains the concept of primary colours and then examines the many colour mod-
els that are available, and the particular strengths and weaknesses of each. A
colour described in one colour model can often be converted to a description in
another. The CIE standard functions as a universal yardstick in this process.

The preceding sections have focused on specifying colour without considering

how colours are physically produced in a computer graphics context. The fourth
section examines how different types of hardware work, emphasising the impact
this has on displaying colour. Guidance is given in making best use of the avail-
able hardware for portable and effective colour graphics.

The final section provides guidelines for using colour. Rather than presenting an
arbitrary series of rules, the intention was to show how the guidelines follow di-
rectly from the material presented in the preceding sections.

There are a three appendices related to displaying standardised colours on a

computer graphics screen, and a glossary of terms. All words in bold like this:

technical term

will be found in the glossary.

University of Manchester 1
2 Seeing in colour

2.1 The electromagnetic spectrum

Light is a form of energy. Visible light is only one form of electromagnetic en-
ergy; other forms include infrared, ultraviolet, radio waves, microwaves and X
rays. Electromagnetic energy can be considered to behave like a wave, and the
factor that distinguishes these many types of energy is the wavelength. This is
illustrated in Figure 1, which uses a logarithmic scale to encompass the wide
range of wavelengths. Visible wavelengths are most conveniently measured in
nanometres (nm, 10-9 m).

Wavelength (m)
4
10
Long wave radio
Ultra violet
2
10 300

VHF radio 0 Violet

10 400
Purple
-2 Blue
Radar 10 500
Green
Yellow
Microwaves -4 Orange
10 600

-6 Red
10 700
Visible light
-8
10 800
X rays Infra red
-10
10

Gamma rays -12

10

Figure 1: The electromagnetic spectrum.

University of Manchester 3
 The ranges of wavelengths which broadly correspond to the colours of the spec-
trum are shown in Table 1 and Plate 25.

Range (nm) Colour

380 – 450 Violet

450 – 490 Blue

490 – 560 Green

560 – 590 Yellow

590 – 640 Orange

640 – 730 Red

Table 1: Approximate wavelengths of spectral colours.

White light consists of a mixture of all the visible wavelengths, which was first
described by Sir Isaac Newton in the Optiks (1704). He found that white light
could be split by a glass prism into a rainbow of colours, and combined again to
form white. He also found that individual colours could not be further subdi-
vided.

2.2 Spectra
It could be imagined that measuring the intensity of light emitted or reflected
from an object at all visible wavelengths would completely define its colour. Such
a measurement will indeed define those optical properties which influence the
observed colour. An example of such a measurement is given in Figure 2. There
is no easy way to predict the visual appearance from this information. The domi-
nant wavelength can readily be identified, but what of the contribution from
the rest of the spectrum? What will the overall colour be?
Relative reflectance

0.2

0.1

0.0
400 500 600 700

Wavelength (nm)

Figure 2: Typical reflectance spectrum of grass.

4 Computer Graphics and Visualisation

 The range of wavelengths which are visible varies between species; some snakes
can see portions of the infrared, and many insects can see into the ultraviolet.
When white light is split by a prism, the wavelengths are separated, but it is the
eye and brain that produce the sensation we call colour.

2.3 The eye

The function of the eye is to capture a visual image, and convert the light energy
into nerve impulses to be interpreted by the brain. The overall structure of the
human eye, shown in Figure 3, is analogous to a camera. Table 2 compares the
functions of the eye and a video camera.

Conjunctiva

Zonula Retina

Aqueous humour
Fovea

Lens

Pupil

Cornea

Iris
Optic nerve

Figure 3: The human eye.

University of Manchester 5
 Eye Video Camera Function

cornea and primary focusing lens bend light to form image

aqueous humour

lens secondary lens fine focusing

iris aperture depth of field & light level adjust

zonula auto focus move lens

conjunctiva clear daylight filter protect optics from scratches

sclera casing mechanical framework

retina photoelectric surface convert light to electrical signal

retinal blood ves- power cables supply energy to retina

sels

optic nerve video signal output transmit data

Table 2: Comparison of the eye with a video camera

The major optical power of the eye comes from the transparent, curved cornea,
which can bend light because of the large change in refractive index between air
on the outside and the liquid (aqueous humour) on the inside. This delicate com-
ponent is covered by the conjunctiva to prevent scratching from small particles
such as grit, dust and smoke; tears are continually secreted to wash the
conjuctiva, and the combination of eyelashes, eyelids and the bony structure of
the skull protect the eye against more major damage.

The iris is a muscle which, when contracted, covers all but a small central por-
tion of the lens, blocking the majority of light and increasing the depth of field.
This provides a greatly increased dynamic range of usable viewing conditions,
from very dim to very bright. The process of responding to a large change in over-
all light intensity is termed adaptation.

Focusing on objects at different distances is accomplished by the lens, which is

moved by a muscle called the zonula. Some of this movement is like a camera,
forwards and backwards. The lens in the eye is however pliable, and can be
pulled at the edges to form a thinner, flatter shape with a longer focal length.
The portion of the lens not covered by the iris looks black from the outside, and is
termed the pupil.

Because the refractive index of the lens and aqueous humour varies with wave-
length, different colours require slightly different lens positions for crisp focus.
This is termed chromatic aberration, and is noticed by a blurring of focus
when colours of widely separated wavelength are seen side by side.

6 Computer Graphics and Visualisation

2.4 The retina
Light energy is transformed to electrical impulses by the retina, a thin network
of cells lining the back and sides of the eye. In some respects, this is like an array
of charge coupled devices similar to that used in video cameras. It is however a
more complex device than this, and it is necessary to have some idea of its action
to understand how colour is perceived.

The cells making up the retina are specialised nerve cells, and are related devel-
opmentally and morphologically to the nervous tissue in the brain. Thus, some of
the retinal nerve cells perform visual processing even before the signals have left
the eye. Another curious consequence of the embryological development of the
eye from brain tissue is that the retina seems ‘inside out’; as Figure 4 shows,
light has to pass through the ‘wiring’ of nerve cells to reach the photosensitive
cells, which are at the back face of the retina.

ganglion cell choroid

horizontal cell rod

bipolar cell cone

Light

Figure 4: The human retina (schematic diagram).

The light sensitive receptor cells at the back of the retina face onto a black lining,
the choroid, which enhances contrast by eliminating internal reflections and pre-
venting light filtering through the front of the eyeball. Receptors are connected
via bipolar cells (so called because of their double-ended shape) to ganglion nerve
fibres, which pass out of the eye to form the optic nerve leading to the brain.

The retina also contains horizontal cells, which connect small clusters of recep-
tors. When a receptor is illuminated, adjacent receptors are made less sensitive
by the horizontal cells, increasing the local contrast. This is a preliminary form of
edge detection, and causes an optical effect known as Mach banding. This is il-
lustrated in Figure 5, which shows a series of grey rectangles. Each is a uniform

University of Manchester 7
 shade of grey, but the edge near the darker rectangle looks lighter. Similarly, the
edge near the lighter rectangle looks darker. Mach banding can be troublesome
in computer graphics; if a smooth gradation in lightness is simulated by a small
number of shades, ensure that the edges are ragged to reduce this effect.

Figure 5: Mach banding.

2.5 Receptor cells

There are two classes of receptor cells: rods and cones, named from their shape.
These are illustrated in Figure 6. They have a similar structure: a central nu-
cleus, many mitochondria to provide chemical energy, and a stack of disks
containing photo-sensitive pigment.

Figure 6: Schematic diagram of human rod (left) and cone (right) cells.

8 Computer Graphics and Visualisation

 Rods are sensitive to very low light levels, but reach their maximum output at
only moderate light intensities. Thereafter they give a constant output regard-
less of increases in light level. Cones are less sensitive, but can handle high light
intensities.

The light sensitive pigment in rods, called rhodopsin, is a protein bound to a form
of vitamin A. Absorption of a single photon of light causes a molecule of rhodop-
sin to change from a low energy to a high energy form. This small energy change
is greatly amplified by a cascade of chemical reactions, to produce a nervous sig-
nal. Unlike most nerve cells, which transmit impulses in a digital, on/off form,
the receptor cells produce a graduated, analogue response to light intensity,
rather like a light meter.

Figure 7 plots the ISO standardised luminous efficiencies of a statistically

normal observer (around 96% of the population; the rest have various forms of
atypical colour vision frequently but incorrectly termed ‘colour blindness’). This
represents the perceived brightness of a light as the wavelength is varied while
holding the light level constant.

At low light levels, when the eye is dark adapted, only the rods are active. This is
termed scotopic vision, and is most sensitive in the green region, at 510nm. In
brighter light, rods are overloaded and the cones are active; the maximum lumi-
nous efficiency for this photopic vision shifts to the yellow/green region at
555nm. This effect is termed the Purkinje shift.

1.0
Scotopic vision
Photopic vision

0.8
Relative luminous efficiency

0.6

0.4

0.2

0.0
300 400 500 600 700 800
Wavelength (nm)

Figure 7: Luminous efficiency.

2.6 Colour reception

In addition to sensing the brighter lights, cones also provide colour sensation.
There are three types of cones, differing in the protein component of the visual

University of Manchester 9
 pigment and thus in the range of wavelengths of light to which they are sensi-
tive. Referred to as S, M and L cones (for short, medium and long wavelengths)
they have maximal sensitivities at 445nm (violet), 535nm (green) and 570nm
(yellow). They are also called β, γ and ρ cones by some authorities.

In contrast to measuring the overall luminous efficiency of a human subject, de-

termining the response of an isolated single type of cone cell is more difficult and
the precise results depend to some extent on the methods and assumptions used.
Two approaches have commonly been used: experiments on subjects who lack
one of the three cone types (which assumes the other two are normal) and meas-
urements of extracted cone pigments (which assumes that the rest of the cone
cell and other retinal structure has no modifying effect).

An example of one set of measured spectral sensitivities for the three cone types
is shown below in Figure 8. It is immediately apparent that S and M cones dis-
play significant overlap and have similar sensitivities and wavelength maxima;
L cones have a much lower sensitivity.

0.8

L cones
M cones
0.6 S cones
Relative sensitivity

0.4

0.2

0.0
400.0 500.0 600.0 700.0
Wavelength (nm)

Figure 8: Sensitivity curves for the three cone types.

Examination of the data on a log scale Figure 9 shows that all three cone types in
fact have similar, low sensitivities in the blue and purple region, but L cones do
not have the large, short-wavelength sensitivity peak possessed by S or M cones.

10 Computer Graphics and Visualisation

 0.0
L cones
M cones
S cones

-1.0

Log relative sensitivity

-2.0

-3.0

-4.0

-5.0
400.0 500.0 600.0 700.0
Wavelength (nm)

Figure 9: Cone sensitivities on a log scale.

One consequence of the marked overlap between M and L cones is that their re-
sponses to a given colour will be highly correlated. Transmitting the signals from
each cone type straight to the visual cortex in the brain would therefore be ineffi-
cient, It would also require four separate signals - S, M, L and brightness. What
happens instead is a current topic of research and debate. All researchers seem
to agree, however, that colour difference signals are produced.

Most researchers agree that in a second stage of colour detection, the difference
of the M and L cones is used to provide a signal which discriminates between or-
ange and bluish green.

The sum of the M and L cones is also transmitted, to provide a brightness chan-
nel. Rods, which are saturated at photopic light levels, provide a small and
effectively constant input to this channel, but there seems to be no input from S
cones.

A second colour difference channel is provided by a weighted combination of S, M

and L cones to aid discrimination of greenish yellow from purplish blue. The ex-
act weightings and combinations of cones, and the probable multiplexing of
difference signals for transmission to the brain, are not yet known with cer-
tainty.

It is likely that there is a third stage of colour processing, somewhere in the

brain, to generate three opposing pairs of colours: black/white. red/green and
blue/yellow. Each of these colours is commonly seen as being in some way unique
or distinct from the others. This idea of opponent colours has a long history,
being described by Leonardo da Vinci and Goëthe, among others.

These three stages – detection by rods and cones, initial combination and final
combination – are shown in Figure 10. The thickness of each line gives an idea of
the contribution of each factor. For example, the contribution of S and M cones to
the second stage greenish yellow / purplish blue channel is about equal, with a
smaller contribution from L cones.

University of Manchester 11
 rod
S M L

+/- - +

+ -
yellow/blue red/green black/white

Figure 10: Opponent colour signals.

2.7 The fovea

This is a small yellowish spot on the retina, directly in line with the optical axis
of the eye shown in Figure 3. The fovea is the area most sensitive to subtle vari-
ation in colour, and is also most sensitive to small details of lightness and shape.
It contains few rods, in contrast to the outer edges of the retina, far from the op-
tical axis, which are rich in rods and optimised for detecting motion at the edges
of vision – a clear survival advantage. Because the eye can be moved to point to-
wards any object of interest, the small size of the fovea is not a disadvantage.

The nerves and blood vessels which cross most of the retina, and through which
light must pass to reach the photoreceptors, are pushed aside from the fovea to
reduce blurring. Extra large M and L cones are packed into the fovea in a hex-
agonal tiled pattern, to give the maximum spatial resolution for the lightness
(M + L) and orange/bluish-green (L – M) second stage channels. Less than two
percent of the foveal receptors are S cones, as these make no contribution to the
lightness channel. A consequence of this low S density is that the spatial resolu-
tion - the amount of fine detail that can be seen - is much lower for blues and
violets than for other colours.

12 Computer Graphics and Visualisation

3 Measuring colour

3.1 Colour matching

As we have seen the phenomenon of colour is a subjective one; a model of colour
must take into account our knowledge of the mechanism of colour vision if it is to
have any utility. The most obvious way to compare two colours for similarity is
thus to look at them side by side. In the early days of colour science, this is ex-
actly what was done.

3.1.1 Tristimulus colour matching

To exactly reproduce the colour of a given object, it would at first seem necessary
to have a multitude of light sources corresponding to all the different spectral
colours and adjust the intensity of each one separately until the objects spectrum
was precisely duplicated.

In practice, however, equivalent colour sensations can be produced by a mixture

of only three colours. This is because the analytical resolving power of the eye for
colour is poor, compared to the resolving power of other sense organs such as the
ear or the nose; a complex light stimulus is perceived as a single sensation. For
example, given a complex audio stimulus, such as a concert orchestra, it is possi-
ble to resolve the individual instruments and listen to just the violins. It is not
possible, given a complex visual stimulus such as white light, to ‘pay attention’ to
just the red components. (In contrast, the spatial resolving power of the eye is
much better than that of the ear).

700

546.1

435.8

700

546.1

435.8

Figure 11: Trichromatic colour matching.

University of Manchester 13
 Figure 11 shows the experimental set up for a colour matching experiment.
Lights are projected onto a diffuser, shown as a grey rectangle in the figure, so
that the observer sees a uniform single colour. The unknown light to be matched,
marked ‘?’, is viewed side by side with the three standard lights, the intensities
of which are individually varied until the colours are seen to match.

Three lights often chosen for colour matching experiments are monochromatic
(single wavelength) sources at 700nm (scarlet red), 546.1nm (yellowish green)
and 435.8nm (bluish violet). These are shown in Figure 11 and Plate 26.

The green and blue-violet lights correspond to sharp peaks in the spectrum from
a mercury vapour lamp; this allows calibration and exchange of experimental
data between different sites. The red light is in an area of the spectrum where
changes in wavelength produce little change in perceived colour, minimising the
effect of mis-calibration.

In some cases, a match can only be obtained by adjusting the unknown colour;
this is done by adding a proportion of one or more of the standard lights using a
second set of standard lamps, shown greyed in Figure 11. This is equivalent to a
negative quantity of one or more lights being required. The specification of a col-
our in terms of the amounts of energy required from each of the three lights to
match it is termed its tristimulus value.

3.1.2 Additivity
The colour resulting from two coloured lights can be exactly predicted; it is the
sum of the tristimulus values of the two lights. The tristimulus value of a 50-50
mixture of two lights is thus the average tristimulus value.

This important property, termed additivity, allows the colour of a mixture of an

arbitrary number of lights to be predicted. Considering the spectrum of a colour
to be made of a large number of wavelength bands allows the tristimulus value of
any object to be calculated as the additive mixture of these bands.

If the predicted tristimulus value of a mixture is positive, it can be mixed with

the colour matching apparatus described. This is true even if one of the compo-
nents of the mixture has negative tristimulus values.

3.2 A standardised observer

Based on a series of matching experiments, a standard observer was defined in
1931 by the International Lighting Committee (Committee Internationale de
l’Éclairage, CIE). This is a set of data which defines three primary colours for col-
our measurements and states, for each wavelength interval, the amount of these
primaries which would be required to match a spectrally pure colour for a statis-
tically normal (non ‘colour blind’) observer. A graph of the matching functions is
shown in Figure 12.

14 Computer Graphics and Visualisation

 2.00

X
1.50 Y

Amount of primary for match

Z

1.00

0.50

0.00
350.0 400.0 450.0 500.0 550.0 600.0 650.0 700.0 750.0
Wavelength (nm)

Figure 12: CIE colour matching functions.

The three primaries are called X, Y and Z. A graph of the amounts of each CIE
primary required to match any pure spectral colour is called the matching func-
tion, and is shown in Figure 12. To match a particular colour, a vertical line is
drawn at that colour’s wavelength and the quantities read off from the intersec-
tions with each matching function. For example, to match the blue/violet colour
of wavelength 450 nm requires 0.33 units of X, 0.04 units of Y and 1.77 units of
Z.

In mathematics, many problems are made easier by the use of imaginary num-
bers, which contain the square root of minus one. Similarly, the X, Y and Z
primaries used to define the standard observer are ‘imaginary’ colours, in that
they do not correspond to visible colours. They have the property of being consid-
erably more saturated than real colours, so that no real colour requires a
negative contribution from any of the primaries for a colour match. X is a super-
saturated purplish red; Y is a supersaturated form of the real spectral green of
wavelength 520nm, and Z is a supersaturated form of the real spectral blue of
wavelength 477nm. Also, the spectral matching function of the Y primary was
chosen to exactly match the CIE standard photopic luminous efficiency function
shown in Figure 7. It therefore carries all the luminance information about the
colour.

The assumptions made in defining this observer were that the colour subtends a
visual angle of 2° or less and the illuminant is not too dissimilar to daylight.
These are easy conditions to meet in practice. The restriction on angular size is
so that the image of patch of colour on the retina falls on the fovea, the area most
sensitive to small changes in colour. This is the usual situation when looking at a
coloured object. For those occasions – rare in practice – where wide field colour
specification is required, the CIE 1964 Supplemental Observer should be used.
The experiments which were performed to define the 1964 observer used a 10°
field, so the resulting CIE values are denoted X10, Y10 and Z10. The shape of the
matching functions is broadly similar to the 1931 Standard Observer, although
tristimulus values calculated from the two observers are different and should not
be mixed.

University of Manchester 15
 Because the standard observer is a mathematically defined set of functions, the
results of colour matching experiments can be simply calculated without actually
having to do the experiment. Colour measurement can thus become an auto-
mated process.

3.3 Coloured objects

There are two broad classes of coloured object, shown in Figure 13 and in Plate
27. Emissive objects produce their own light. Reflective objects, on the other
hand, are totally dependent on an external light source; they provide a modifica-
tion of the colour of the illuminant by absorbing different amounts of light at
different wavelengths. Fluorescent objects are a special case, in that they take in
light at one wavelength and re-emit some of it at a longer, lower energy wave-
length.

Surface gloss
Illuminant Emitted colour
reflected colour

Light source

Figure 13: Reflective and emissive objects.

3.4 Emissive colour

3.4.1 Calculating tristimulus values

The colour of a light source is defined by the quantities of X, Y and Z primaries
which would be required to match it. To calculate these quantities, the visible
spectrum of 380 to 730nm is divided into a number of wavelength intervals and
the intensity of the sample measured for each interval to produce its spectrum. A
10nm interval is commonly used. Then, for each primary in turn, the height of
the sample spectrum is multiplied by the height of that primary’s matching func-
tion. These products are summed across all wavelength intervals to calculate the
overall quantity of that primary required for the match. This is illustrated in
Figure 14.

16 Computer Graphics and Visualisation

 x x x

= = =

Σ Σ Σ

X Y Z
Figure 14: Calculation of CIE tristimulus values.

3.4.2 The XYZ diagram.

The resulting value (X, Y, Z) may be plotted on a 3D diagram, and will fall in the
positive (XYZ) quadrant within the cone-shaped solid shown in Figure 15 and in
colour in Plate 28. Notice that the coordinate axes are not inside this solid; the
XYZ primaries are imaginary colours. Black, corresponding to the lack of light, is
at the origin. The curved boundary represents the tristimulus values of pure
spectral colours. Because these are of a single wavelength, they represent the
maximum attainable saturation. This boundary is called the spectral locus; all
visible colours are inside or on it.

Wavelengths from 400nm (violet) to 700nm (red) are shown; note that the wave-
length spacing is not at all even. The straight line connecting the ends of the
spectral locus corresponds to additive mixtures of the red nearest infrared and
the violet nearest ultraviolet to produce purple.

University of Manchester 17
 Y
520nm

500nm 600nm

700nm

Z 400nm

Figure 15: The CIE 1931 XYZ diagram.

3.4.3 Automatic measurement

In practice, the process of measuring the spectrum of a light source and obtain-
ing its tristimulus values is automated. An instrument called a
spectroradiometer measures the brightness at each wavelength interval with
a photocell, and has the values of the standard observer matching functions at
each wavelength interval stored internally. A small microprocessor performs the
calculations, and readout is directly in the form of X Y and Z tristimulus values.

A luminance meter is similar but cheaper, as it only gives a readout of the Y

measurement.

3.5 Illuminants
Many objects are coloured which do not themselves emit light. The colour is due
to reflection of light from other light sources. The object absorbs varying amounts
of light of each wavelength, and the unabsorbed portion is reflected back to the
eye to give the sensation of colour. Clearly, the quality of light illuminating the
object will affect the perceived colour.

18 Computer Graphics and Visualisation

3.5.1 White light
What is seen as white light generally does not have a flat spectral response. Hu-
man eyes have evolved to consider daylight the ‘normal’ illuminant. For
centuries, artists have used overcast northern daylight for preference when mix-
ing colours, an important consideration when a painting is being worked on for a
prolonged time and colours must match those used earlier in the work. The qual-
ity of daylight is very variable, depending on location, season and degree of cloud
among other factors. It is clearly desirable to have a standardised light source,
representative of daylight. It is no surprise then that the CIE has recommended
that daylight-like sources be used for matching.

3.5.2 Illuminant C
The CIE have produced such a standard, called illuminant C. This consists of two
parts:

1. a tungsten light, the electrical filament of which is operated at a precise

temperature, which gives a very warm, orange light. (This light is a stan-
dard, Illuminant A, but is little used by itself.)

2. a filter, consisting of a tank of blue liquid, the chemical composition of

which is specified by the CIE, to produce a simulation of neutral daylight.

While this arrangement is suitable for laboratory use, the light output is fairly
low and the liquid tank inconvenient. Furthermore, Illuminant C is lacking in
the near ultraviolet region, making it less useful for fluorescent materials. Rec-
ognising this, the CIE has specified another class of daylight illuminants which
rectify these deficiencies.

3.5.3 The D series illuminants

These differ according to their colour temperature, which is a one dimensional
specification for the colour of an approximately white light; the colour produced
by a theoretical black body radiator when heated to a particular white-hot tem-
perature. The most common D illuminant corresponds to a temperature of 6500°
Kelvin, and is thus termed D65. Other illuminants in the D series are sometimes
encountered, notably D50 (5000°K) in the graphic arts industry. The spectra of
illuminants A, C and D65 is shown in Figure 16. Notice that in the ultraviolet
region (below 400 nm) the curve for Illuminant C is much lower than for D65.

University of Manchester 19
 Illuminant A
Illuminant C
Illuminant D65
2.0

Relative intensity
1.0

0.0
300.0 400.0 500.0 600.0 700.0 800.0
Wavelength (nm)

Figure 16: Spectra of some standard illuminants.

The spectrum of D65 is complex, and difficult to reproduce exactly with an artifi-
cial light. There is no actual apparatus specified as part of the standard. Instead,
there is a procedure to determine how good a particular light is at giving the
same colour matching results as D65.

A number of selected paint samples are measured under the light in question
and their tristimulus values compared to theoretical values calculated from the
reflectance spectrum of each paint chip and the spectrum of D65. The results
from each chip are combined to obtain an overall index of colour matching fidel-
ity. Over 90% is a good result, below 60% will give significant errors.

The phenomenon of two samples matching in colour under one illuminant but
not another is called metamerism. The CIE procedure is designed to minimise
metamerism between the theoretical D65 and practical implementations of it.

3.5.4 Other illuminants

Fluorescent strip lamps are frequently encountered as indoor lighting. These
typically produce metamerism with colours which were specified using a daylight
illuminant. Figure 17 compares the spectrum of a typical fluorescent light with
D65. The differences are considerable.

20 Computer Graphics and Visualisation

 6.0

Illuminant D65
Fluorescent striplight

4.0

Relative intensity

2.0

0.0
300.0 400.0 500.0 600.0 700.0 800.0
Wavelength (nm)

Figure 17: Spectra of daylight and a fluorescent striplight.

Ordinary household light bulbs have a spectral distribution similar in overall

shape to Illuminant A, shown in Figure 16. As this is significantly different from
D65, metamerism problems can be expected with light bulbs too. It is unfortu-
nate that the two most common sources of indoor illumination are very different
from daylight. This is why, when a precise estimate of colour quality or colour
balance must be done visually, a special matching cabinet is used. This is simply
a box containing a standard illuminant, into which the sample is placed for view-
ing. It is painted grey on the inside to provide a neutral background.

3.6 Reflective colour

Measurement of the tristimulus values of a light source involves multiplying the
emission spectrum of the sample with the spectra of the three standard observer
matching functions. For reflective light, the procedure is the same except that
the emission spectrum of the illuminant is first multiplied by the percentage re-
flectance of the sample for each wavelength interval. This converts the spectrum
of the incident illuminant to the spectrum of the reflected light. For this reason,
specification of the colour of a reflective object by tristimulus values is meaning-
less without also specifying which illuminant was used for the measurement.

3.6.1 Adaptation
The visual appearance of a white surface, such as a piece of paper, varies slightly
when seen under different illuminants, but still looks white. In contrast, the

University of Manchester 21
 measured spectrum and tristimulus values will vary considerably. This phe-
nomenon is termed colour constancy and is due to complex visual processing
in the brain which tries to correct for slow changes in overall light level and qual-
ity. This is dependent on the overall state of adaptation of the eye. For example,
a white sheet of paper will look white under the orangish glow of a tungsten light
bulb or the bluish glare of a fluorescent striplight. This is because the illuminant
fills the whole field of vision and provides the dominant adaptive stimulus. If a
photograph is taken of the paper under tungsten light, there will be a marked
orange tint to the white paper when the print is developed. This is because the
amount of light reflected from the photograph is a small fraction of the total light
entering the eye, which does not therefore adapt to it.

3.6.2 The white point

To account for this phenomenon, it is customary to define a white point which is
taken to be the colour currently accepted as white. For emissive colour, this is
one of the standard ‘white’ illuminants. For reflective colour, a theoretical entity
called the perfect diffuse reflector is defined to have the convenient property
of 100% reflectance at all wavelengths, which allows the illuminant to become
the white point.

The tristimulus values of a reflective sample may be measured with a

spectrophotometer, which is similar to a spectroradiometer except that it also
contains a light source. A spectrophotometer is calibrated by simply measuring a
pure white surface, to reflect the illuminant back onto the detector.

3.6.3 Fluorescent colours

These have the property of absorbing light at one wavelength and re-emitting it
at a longer wavelength (which will have less energy). In many cases, the ab-
sorbed light will be in the ultraviolet range but the emitted light will be visible.
As this affects the measured tristimulus values, it is important that a daylight
illuminant is used which has similar levels of ultraviolet to real sunlight, a fail-
ing of the original Illuminant C.

Substances which absorb ultraviolet and emit blue light are used in some wash-
ing powders; the blue neutralises the yellowish tint of residual soiling to give the
illusion of white. Such substances are termed optical brighteners.

3.7 Chromaticity

3.7.1 The CIE 1931 (xy) diagram

If a given colour is increased in brightness, the amount of light required from
each primary to match the colour increases. The increases will be in proportion,
so the ratio X:Y:Z remains constant as the colour moves away from the origin.

It is often useful to examine the colour of a sample separately from its bright-
ness. To do this, the tristimulus values are normalised:

22 Computer Graphics and Visualisation

 x = X / (X + Y + Z)

y = Y / (X + Y + Z)

z = Z / (X + Y + Z)

Clearly, x + y + z = 1 in all cases. It is therefore customary to drop the z

coordinate and produce a 2D plot of x against y. This is equivalent to projecting
the XYZ colour solid onto the X + Y + Z = 1 plane, which is shown in Figure 18.

The X+Y+Z=1 plane

Z
Figure 18: The X + Y + Z = 1 plane on the CIE 1931 XYZ diagram.

The resulting diagram is called the CIE 1931 chromaticity diagram, and is
shown in Figure 19 and in Plate 29. It represents the perceptual attributes of
hue and saturation, separated from luminance.

Chromaticity (x,y) values are sometimes encountered together with a Y value

(xyY) to allow conversion back to XYZ.

University of Manchester 23
 525
0.8

550

0.6
500 575

y 0.4 600

660-780

0.2

475
450
0.0
0.0 0.2 0.4 0.6 0.8
x
Figure 19: The CIE 1931 chromaticity diagram

Remember that in Section 3.1 (page 13) colour matching experiments used spe-
cific calibrated wavelengths for yellow/green and blue/violet lights, but red light
was chosen to be in a wavelength range where mis-calibration would not affect
the colour. Looking at Figure 19, notice that colours from 660nm (red) right up to
780nm, the limit of visible colour bordering on infra-red, are at the same place on
the chromaticity diagram; they are all the same hue. A similar ‘bunching up’ of
wavelengths occurs at the border of violet and ultraviolet (380nm).

One problem with this diagram is that it is not perceptually uniform. In other
words, the distance between two colours which are just noticeably different var-
ies across the surface of the diagram.

3.7.2 CIE 1976 uniform chromaticity scale (UCS)

Investigation of the perceptual uniformity of the 1931 chromaticity diagram, by
plotting the size of a just noticeable colour change in various parts of the dia-
gram showed that the portion at the top of the curve, in the green region, showed
little variation in colour compared to the blue violet portion. This perceptual dis-
crepancy has been likened to the distortions on flat maps of the world, in that it
cannot be removed entirely by a linear projection of the 1931 chromaticity dia-
gram. However, some projections will be better than others. One such projection
was recommended by the CIE in 1976, the uniform chromaticity scale (UCS).

Also known as the u′v′ diagram, from the labels on its axes, this is a projection of
CIE 1931 XYZ space designed to produce much less distortion than the xy dia-
gram. (The curious ‘dash’ notation is intended to distinguish it from a similar
(uv) diagram which was used prior to 1976.) The axes are calculated as follows:

u′ = 4X/(X + 15Y + 3Z) = 4x/(-2x + 12y + 3)

24 Computer Graphics and Visualisation

 v′ = 9Y/(X + 15Y + 3Z) = 9y/(-2x + 12y + 3)

The resulting diagram, shown below in Figure 20, is of the same general shape
as the 1931 chromaticity diagram but stretched to give a more uniform distribu-
tion of colours.

0.6 525 550

575
600 675-780
500

v’ 0.4

0.2 475

450
400
0.0
0.0 0.2 0.4 0.6
u’
Figure 20: The CIE 1976 uniform chromaticity scale diagram

3.7.3 Putting chromaticity diagrams to work

Chromaticity diagrams have a variety of uses. They all share the property that
an additive mixture of two colours will lie along the line connecting them. This
can be used to calculate colorimetric data such as the dominant wavelength
and excitation purity of a colour.

Dominant wavelength is determined by constructing a line from the white point,

through the colour in question, to the spectral locus and reading off the wave-
length. This is shown in Figure21. Regardless of the spectral composition of the
original colour, an equivalent colour sensation will be obtained by mixing mono-
chromatic light of the dominant wavelength with the specified white. The
excitation purity gives the proportions of this mixture. In this example, the exci-
tation purity is 40%.

University of Manchester 25
 525
0.8 dominant wavelength 540 nm
550

0.6
500 575
sample

y 0.4 600

D65 whitepoint
660-780

0.2

475
450
0.0
0.0 0.2 0.4 0.6 0.8
x
Figure 21: Dominant wavelength calculation

In other words, the sample colour is made from the additive mixture of 40% spec-
tral light of wavelength 540nm, and 60% of D65 white at the same luminance.

Chromaticity diagrams may be used to pick complementary colours by selecting

points opposite to each other across the white point, and to design continuous col-
our scales by tracing straight or curved paths across the diagram. In general,
however, a three dimensional colour model is used for this task so that the
brightness of the colours can also be altered. Colour models are discussed in the
next section.

3.8 Atypical colour response

More than 95% of the world population have statistically normal colour vision.
These are the people whose colour matching abilities are represented mathe-
matically by the CIE standard and supplemental observers.

Although people with an atypical colour response are often called ‘colour blind’,
there are in fact very few people who have absolutely no colour perception. The
proportion of the population with reduced colour discrimination is however sur-
prisingly large. Atypical colour response is sex-linked; while being found quite
rarely in Caucasian females and non Caucasians, it occurs in 8% of Caucasian
males.

People with normal colour vision are termed trichromats because they have three
functional cone types. Those with reduced functioning of one cone type are called
anomalous trichromats.

26 Computer Graphics and Visualisation

In some people, one cone type is absent or completely non-functional. Because
they have two functioning cone types, they are termed dichromats. Finally, those
very rare people with only one functional cone type, who see only in shades of
grey, are called monochromats.

Dichromats will see certain colours as the same which are clearly different to tri-
chromats. These confused colours lie along straight lines on a chromaticity
diagram, converging on a single point.

As one would expect, there are three types of dichromatism, depending on which
one of the three cone types is affected. Each type has a different confusion point.

• Protanopia is due to missing or dysfunctional L cones. Protanopes have

greatly reduced discrimination of reds from greens, and reds look dimmer
than normal. The confusion point P, shown in Figure 22, is at or near u′ =
0.61, v′ = 0.51, very close to the far red corner of the chromaticity diagram.

0.6 525 550

575
600
500 P

675-780
0.4

v’

0.2
475

450 400
0.0
0.0 0.2 0.4 0.6
u’

Figure 22: Protanopic confusion lines.

• Deuteranopia is due to missing or dysfunctional M cones. It can also be

caused by the lack of the L-M colour difference signal shown in Figure 10.
Deuteranopes also have reduced discrimination of reds from greens, but
without any colours seeming dimmer than normal. The dueteranopic confu-
sion point lies well off to the left of the chromaticity diagram, at or near u′ =
-4.75, v′ = 1.31. As can be seen from Figure 23, this means that the lines of
confused colours are nearly parallel.

University of Manchester 27
 0.6 525 550
575
600
500

675-780
0.4

v’

0.2
475 D at
u’ = -4.75
v’ = 1.31

450 400
0.0
0.0 0.2 0.4 0.6
u’
Figure 23: Deuteranopic confusion lines.

• Tritanopia is caused by a lack or deficiency in the S cones. Because these

make no contribution to the lightness channel, all colours are the same
brightness as normal. However there is greatly reduced discrimination be-
tween yellows and blues. The confusion point, shown in Figure 24 is at or
near u′ = 0.26, v′ = 0.003, very close to the violet corner on the chromaticity
diagram.

0.6 525 550

575
600
500

675-780
0.4

v’

0.2
475

450
0.0
0.0 0.2 T 0.4 0.6
u’
Figure 24: Tritanopic confusion lines.

28 Computer Graphics and Visualisation

In each case, confused colours lie along converging lines; these may be used to
select colour schemes that can be used by colour deficient subjects.

University of Manchester 29
4 Colour models

4.1 Why use colour models?

Given the complexities of colour perception, it is useful to define a simplified, ab-
stract method of succinctly specifying colour with a small number of parameters.
Typically, the tristimulus theory is used so there are three parameters. There
may be other underlying assumptions too. Often colour models are defined in
terms of three primary colours, from which all others are obtained by mixture.
In other cases, the three parameters represent more readily understood attrib-
utes such as lightness or saturation.

Considering the three parameters to be orthogonal axes produces a geometric

colour space. The geometric position of a colour in this space can be used to see
its relationship with other colours. 48 shows a program which can be used to mix
colours in a variety of different colour spaces.

4.2 Primary colours

The choice of ‘primary’ colours depends on a number of factors:

• can they take negative values ?

• are they hardware dependent ?

If the primaries can be negative and are not constrained by hardware, any pri-
maries can be used. A single primary is represented by a point on a chromaticity
diagram; it can only produce that one colour at varying intensities. Two prima-
ries produce a line segment on a chromaticity diagram, as shown in Figure 25.
Any colour on that segment can be produced by a non-negative mixture of the
two primaries. If negative values for primary colours are allowed, the line seg-
ment can be extended out from each primary to the spectral locus as shown by
the dotted line in Figure 25; any colour on this line can be specified.

University of Manchester 31
 0.6 525 550
575
600
675-780
500

0.4
B

v’

0.2 475

450
400
0.0

0.0 0.2 u’ 0.4 0.6

Figure 25: Two primaries define a line.

Figure 26 shows how a third primary C can be used to cover the entire visible
area; on the line joining A and B the contribution from C is zero; on the side to-
wards C, the value of C is positive and on the other side it is negative.

0.6 525 550

575
600
675-780
500

0.4
B

C
ne
C ga
v’
tiv
e
C
0.2
ze
475 ro

C
po
sit
450 iv
e
400
0.0

0.0 0.2 u’ 0.4 0.6

Figure 26: Negative and positive contributions from one primary.

Three primaries define a plane, and so any visible colour can be specified with
any arbitrary choice of primaries provided negative values are allowed. Figure 27
shows three random primaries - a yellow (A), a purple (B) and a turquoise (C)
can be used in this fashion. Note the large areas that require one or more prima-
ries to be negative.

32 Computer Graphics and Visualisation

 0.6 525 550
575
600
B- C- 675-780
500

B-
A C-

0.4
All + B
A-C-
C
v’
A-B-

0.2 475 A-

450
400
0.0

0.0 0.2 u’ 0.4 0.6

Figure 27: Three primaries define a plane.

If primaries cannot take negative values, only colours within the central triangle
can be produced. In this case, it makes sense to maximise the area of this trian-
gle by aligning it with the broadly triangular shape of the chromaticity diagram.
This gives one primary somewhere near the far red corner, one in the green cor-
ner and one near the far violet corner.

Another way to increase the range of colours is to use more primaries. Figure 28
shows a system with five primaries. All colours within the pentagon can be pro-
duced in more than one way. Removing any one primary reduces the number of
colours that can be produced.

0.6 525 550

575
600
500 B 675-780
A

0.4 C

E
v’

D
475
0.2

450

400
0.0

0.0 0.2 u’ 0.4 0.6

Figure 28: A system using five primaries.

University of Manchester 33
4.3 CIE colour models

4.3.1 CIE 1931 XYZ

This colour space has already been discussed; it is the primary colour space for
colorimetric measurement and the basis for other CIE colour spaces. It has the
advantage of being rigorously defined, and an international standard. Whilst
good for presenting a description of an existing, measured colour, it is not par-
ticularly easy to use for specifying new colours. This is because the primaries are
not visible colours; while they are always positive, the axes lie outside the range
of visible colour as can be seen in Figure 29.

Y
520nm

500nm 600nm

700nm

Z 400nm

Figure 29: The CIE 1931 XYZ colour space

4.3.2 1976 CIELUV

If the luminance of a colour is divided by that of some reference white, a relative
luminance scale from 0 to 100% (black to white) is obtained. Measured lumi-
nance, however, does not correspond well to perceived lightness; the scale looks
markedly non uniform, with all the dark colours bunched up at one end. The CIE
has recommended a non-linear formula for lightness, L*, which corresponds
more closely to the perceived sensation. The medium, 50% grey occurs at L* = 50.
The appearance of grey scales constructed from luminance and lightness is
shown in the diagram below.

34 Computer Graphics and Visualisation

 Figure 30: Lightness (above) and luminance (below).

L* = 116 (Y/Yw)
1/3
– 16 for most values of Y (Y/Yw > 0.008856) or

L* = 903.3 (Y/Yw)for very dark colours (Y/Yw ≤ 0.008856)

Yw is the Y tristimulus value for the reference white, in other words that white
to which the eye is adapted. For most purposes, a standard illuminant such as
D65 is used as the reference white.

The CIE has recommended a uniform, 3 dimensional colour space incorporating

both the UCS and this lightness parameter, called the CIE 1976 (L*u*v*) colour
space. Shown in Figure 31 and in Plate 30, it is commonly referred to as
CIELUV, and may be considered a uniform version of the CIE 1931 XYZ space.

The formulae are:

u* = 13 L* (u’ - u’w )

v* = 13 L* (v’ - v’w )

where u’w and v’w are the UCS coordinates of the chosen reference white.

These formulae correspond to translating the origin of the UCS diagram to the
white point and scaling the relative chromaticity co-ordinates by the lightness so
that the geometrical distance between two colours is reduced as they are made
darker. This takes account of the fact that dark colours look more alike than
light ones, even when the chromaticities are the same. The resulting colour space
therefore forms a cone-like solid. Black, at L* = 0 is thus a single colour at the
apex of this cone.

CIELUV space could also be considered as a stack of scaled UCS diagrams, as

Figure 31 shows.

University of Manchester 35
 L*

-600

u*
-200 200 500
v*
Figure 31: The CIE 1976 (L*u*v*) colour space.

As the spectral colours form a loop around the origin, it is possible to define a
hue angle huv which specifies hue with a single numerical value. The positive u*
axis is defined to be 0°, and angles are measured anticlockwise.

huv = arctan (v* / u*)

An advantage of this is that hue is a readily understood concept. The colours of

the rainbow are arranged in a circle. The distance from the achromatic L* axis
may then be used as a measure of chroma, or colourfullness:

C*uv = (u*2 + v*2)1/2

Lightness, chroma and hue angle define an alternative, polar form of CIELUV,
shown in Figure 32. This is easier to use for mixing colours than CIELUV.

36 Computer Graphics and Visualisation

 L*

C*

v* H

u*

Figure 32: Hue angle and chroma.

Saturation is calculated from:

S* = C* / L*

Figure 33 shows planes of constant chroma and constant saturation in CIE LCH
space. Chroma is clearly seen as independent of lightness.

L*

constant
saturation

v*

constant chroma

u*

Figure 33: Hue angle and chroma.

The CIELUV model, because of its greater perceptual uniformity than other
models, has been used in the film and TV industries and is finding increasing use
in computer graphics. It is generally used for emissive colours such as lights or
computer monitors. Conforming PHIGS and PHIGS PLUS implementations are
required to support this model.

University of Manchester 37
 4.3.3 1976 CIELAB
An alternative colour model is recommended by the CIE for reflective colours
such as paints or dyed fabrics. It is optimised for quantifying the colour differ-
ence between two samples of nearly identical colour, such as between two
batches of dyestuff, and producing similar numerical results to other existing col-
our difference formulae. The lightness parameter is the same L* as in CIELUV,
but the other two are:

1/3
a* = 500 [ (X/Xw)1/3 - (Y/Yw) ]

1/3
b* = 500 [ (Y/Yw)1/3 - (Z/Zw) ]

Because of the cube roots in these equations, there is no chromaticity diagram for
CIELAB. Straight lines in CIE 1931 xy space remain straight in the UCS, but
would not in a chromaticity diagram based on CIELAB. It would therefore not
represent additive mixture.

Since 1976, more complex non-Euclidean formulae have been devised for colour
difference in CIELAB which add varying weights of lightness, hue angle and
chroma depending on the region of colour space in which the samples lie. These
formulae are used in high precision industrial applications

Because reflective colours are being specified or measured, the illuminant used
should also be stated. In general, D65 is used unless there is some special reason
to select another illuminant. Both CIELUV and CIELAB assume that the
illuminant, or reference white, is close to natural daylight.

CIELAB has a polar form called L*C*abhab, the subscripts serving to distinguish
it from the polar form of CIELUV. The formulae are identical.

CIELAB is sometimes encountered in the specification of colour printers. Given

the reference illuminant used, it is simple to convert these values to XYZ or
LUV. The second edition of the Computer Graphics Metafile (ISO/IEC 8632:
1992) allows colours to be specified in either CIELUV or CIELAB. The Open
Document Architecture, an ISO standard for compound documents, uses
CIELAB for colour specification, as does the Image Processing and Interchange
(IPI) standard.

4.4 Device dependent models

It is sometimes convenient or customary to specify colour directly in the native
colour space of a particular device. In the case of devices which use emitted light,
such as colour monitors, an additive geometrical space can be produced subject to
certain constraints discussed later (Section 5 ). Devices which use reflected light,
such as all forms of printing, are not additive and a geometrical space cannot be
formed. While individual colours can be specified in such a system, the relation-
ship between colours or the result of mixing two colours can only be determined
on an empirical basis.

38 Computer Graphics and Visualisation

4.4.1 Red, green, blue (RGB)
This colour space is commonly used, and corresponds to the input data for a spe-
cific colour CRT computer monitor. The three primaries are the particular
colours emitted by the three phosphors. It is therefore highly device specific; the
same colour will be specified as two different sets of numbers on two different
monitors. The parameters are the quantities of red, green and blue light to emit,
generally in the range 0 to 1. The RGB colour space is shown in Figure 34 and in
Plate 31.

One strength of the RGB colour space is that it is a unit cube, and thus all possi-
ble values of R,G,B correspond to realisable colour. This makes it convenient
from a programming point of view, in that range checking is straightforward.

Red

Magenta Yellow

White

Black

Blue Green

Cyan

Figure 34: RGB colour space.

A major weakness is that colours specified in RGB space are not at all
perceptually uniform, and it is not sensible to measure colour differences in RGB
space. This colour model is further discussed in Section 5.1, (Displaying colour).

If the chromaticity coordinates of the monitor phosphors are known, and also
both the chromaticity and luminance of the white produced by equal quantities of
red, green and blue, it is possible to interconvert between RGB and the CIE col-
our spaces. This conversion is described in Appendix B.

RGB colour space is widely used in computer graphics and is supported by GKS,
PHIGS and most other graphics systems. It is adequate for use in situations
where producing different colours is more important than portability or repro-
ducibility. Specifying colour in RGB space is more convenient if hue, chroma and

University of Manchester 39
 lightness are separate parameters. There are two transformations of RGB space
used to achieve this: HLS and HSV. These will be considered next.

4.4.2 Hue, saturation, value (HSV)

The major diagonal of the RGB cube, from black at (0,0,0) to white at (1,1,1)
forms an achromatic axis or gray scale. If the cube is rotated so that the white
corner points towards the viewer and the black corner points away, as in Figure
34, a hexagon is seen with hues radiating around the achromatic axis. The HSV
colour space uses this concept to define a hue angle, saturation and a third pa-
rameter, value, which broadly corresponds to lightness. HSV is thus a polar
coordinate model, and these terms are analogous to, but not identical with, the
similarly named terms in the polar coordinate L*C*Huv form of the CIELUV
model.

The space, shown in Figure 35 and in Plate 32, is a cylinder centred on the achro-
matic axis. Value is the distance along this axis, saturation is the radial distance
from it. Hue is an angular measure with 0 representing red and 180 representing
cyan (a greenish blue).

Because HSV uses saturation rather than chroma, the perceived change in colour
as saturation varies between 0 and 1 is less for dark (low value) colours than for
light (high value) colours. To compensate for this, the HSV colour space is often
shown distorted to form a cone rather than a cylinder. Other diagrams show
HSV as a hexcone, to reinforce the link with RGB. However, saturation still
ranges from 0 to 1 regardless of value or hue so these changes do not represent
the geometric space accurately.

Green Yellow
White
Cyan Red

Blue Magenta

S H

Figure 35: The HSV colour space

40 Computer Graphics and Visualisation

Being a transformation of RGB space, HSV shares the advantage that all possi-
ble values of H,S and V correspond to displayable colours. In addition, it is easier
to mix colours than in RGB because the three parameters correspond more
closely to perceptual attributes. On the other hand, HSV is just as device specific
as RGB so descriptions of colours in HSV are not portable. Like RGB, only dis-
playable colours can be specified.

Unlike CIELUV, HSV space is not perceptually uniform. Equal increments of

hue angle do not produce smooth changes of perceived hue. Also, the three pa-
rameters are not independent. For example, a pure yellow and a pure blue both
have S=1, V=1 yet the yellow will have a significantly higher luminance and be
perceptually lighter than the blue.

4.4.3 Hue, lightness, saturation (HLS)

Similar to HSV, this colour model has a lightness axis rather than a value axis.
Pure colours, shown in Figure 36 and in Plate 33, have a saturation of 0.5, rather
than 1.0 in HSV. HLS may be considered a simple deformation of HSV produced
by moving the white point as far above the pure colours as the black point is be-
low them. Like HSV, it is a cylindrical colour space but is often drawn as a cone -
in this case, a double ended one.

White

S H

L
Green Yellow

Cyan Red

Blue Magenta

Black

Figure 36: The HLS colour space

HLS, like HSV, is simply another representation of RGB space. These 2 colour
models may optionally be supported in a PHIGS implementation – they are de-
fined but a conforming implementation need not support them.

University of Manchester 41
 4.4.4 Cyan, Magenta and Yellow (CMY)
CMY is sometimes presented as a colour space, and corresponds to the input
data for colour printing. However it deals with the proportions of real pigments
rather than abstract colours. Furthermore, mixing two colours is not additive,
which makes the representation of CMY as a geometric solid of little value.
Specification of colours in CMY, even when the CIE tristimulus values of the
inks are known, is complicated by a great many factors as will be seen in Section
5.5. A variation of CMY adds black ink, and is called CMYK. (Black is referred to
as K rather than B to avoid confusion with blue in RGB).

4.4.5 Video and broadcast colour models

These are derivations of particular calibrated RGB colour spaces, using monitor
primaries whose chromaticities are well defined. In the context of computer
graphics, they are met with when animated sequences are stored on video tape
or broadcast on television. These are hardware-led colour models, optimised to
make best use of the information capacity of the medium, and will be discussed
in more depth in Section 5.1.

4.5 Other colour models

There are a wide variety of colour models which may occasionally be encountered
in computer graphics and visualisation work. Some of these are used in a par-
ticular application domain, such as architecture or textiles. A number of models
are specified as national standards in particular countries and will be met when
producing work for those countries. Others are included here because they illus-
trate a particular point about colour science.

4.5.1 Munsell system

This is a perceptually uniform system for reflective colours used in graphic arts,
textile and paint industries, particularly in the USA. It was recommended by
ANSI, the American National Standard Association. It consists of a book of
painted colour samples classified by three parameters – Munsell hue, value and
chroma. There are five principal hues: red, yellow, green, blue and purple. A fur-
ther five hues are mixtures of these: green/blue, blue/purple. Hues are referred to
by their initial letter, for example R, YG. Each of these ten hues are further sub-
divided by a decimal number. There are four divisions reproduced in the Munsell
Book of Colour (2.5, 5, 7.5, 10), giving 40 hues in total, which radiate in a circle
from the achromatic value axis as shown in Figure 37 and Plate 35.

42 Computer Graphics and Visualisation

 Red
R

Y
P/

R/
le
Purp w
Yello

B/P
Y/G

Blu

G
re
e

G/B

en
Figure 37: The Munsell hue wheel

Value is specified by an integer greater than 0 (black) and less than 10 (white).
Chroma is the radial distance from the achromatic axis, and ranges from 2 to 14
or more; the book has samples at steps of two. All colours with a chroma of 0 are
on the achromatic axis and do not have a hue. An example of a colour specified in
the Munsell system is 5.0PB/4/10 which means a purple-blue with value 4
(darker than a mid grey) and chroma 10 (very colourful, near the maximum at-
tainable pigment limit for that colour). CIE values for all the patches, measured
using Illuminant C, have been published; this particular colour has chromaticity
values of x=0.1773, y=0.1659, and Y=0.1200 for example.

Designed by the artist Albert Munsell, this system relies on the subjective selec-
tion of hues for perceptual uniformity rather than a colorimetric approach. The
original selection of colours was entirely ‘by eye’. Comparisons of the Munsell
system with CIELUV and CIELAB show that, while none of these systems is
completely perceptually uniform, the agreement between them is surprisingly
close.

Colours in the Munsell system correctly separate out chroma and value. For ex-
ample, a medium yellow such as 5.0Y has maximum chroma at a high value; this
corresponds to the observation that the purest intense yellow is a light colour.

The diagram below is intended to give the idea of the Munsell system, and de-
picts two pages arranged about the achromatic axis. It is also shown in colour
Plate 34.

University of Manchester 43
 5PB 5Y

Value

Chroma
Figure 38: A pair of leaves from the Munsell system

4.5.2 Natural Colour System

The natural colour system (NCS) is an opponent colour system recommended
by the Swedish Standards Institution. It uses the theory of opponent colours ad-
vanced by Herring, which states there are two pairs of colours (red/green and
yellow/blue). None of these colours resembles any of the others, and any colour
can only contain a contribution from one of each pair. For example, a colour can
be greenish blue, but cannot be reddish green or bluish yellow. There is a third
pair, black/white, and colours can contain a mixture of both to form greys.

Initially, the opponent theory was disputed by proponents of the trichromatic

theory, who pointed out that a yellow sensation can be produced with red and
green light. It is now generally accepted that opponent colours, which as a con-
cept date back to Leonardo da Vinci’s four visual primaries, are not in conflict
with the trichromatic theory and indeed can be related to the opponent wiring of
the retina. The point is not that yellow can be produced by mixture, but that it
has the visual appearance of a distinct colour, having neither reddish or greenish
tint.

The three parameters of NCS space are hue (φ) blackness (s, from the German
Schwarz) and chromaticness (c). Hues are specified as a percentage of two of the
four colours. For example, a lime green might consist of a mixture of 65% yellow
and 35% green. This would be written as Y35G. The relationship of all colours of
a given hue, φ to the achromatic axis can be drawn on an equilateral triangle.
The distance of a particular colour from each edge may be expressed as a per-
centage; the percentages of white (W) black (S) and the hue (F) will always add
to 100, so it is customary to omit the percentage of white. NCS is shown in Fig-
ure 39 and Plate 36.

44 Computer Graphics and Visualisation

 R
W

φ
B Y

Blackness
Y35G

S
Chromaticness G

Figure 39: The Swedish NCS

Plotting the NCS colours on a CIE UCS shows that chromaticness is fairly uni-
form, but hue is not; there are more colours in the blue to red quadrant than the
others. This means that complementary colours selected with NCS will not be
the same as those selected with CIELUV or Munsell colour models.

NCS has been used in some paint catalogues for the painting and decorating
trade. Accordingly, it may occasionally be encountered in architectural visualisa-
tion work.

4.5.3 DIN system

The German national standards body, Deutsches Institut für Normung (DIN)
has developed a colour model (DIN 6164) which again has a circular hue specifi-
cation. Colours of the same dominant wavelength have the same hue (T), which
ranges from 1 for yellow through red, purple, blue and green to 24 for greenish
yellow. The other two variables are saturation (S) and darkness (D). The overall
shape approximates a round-topped cone, with black at the apex. However, the
maximum saturation varies for different hues. Because of the use of saturation
rather than chroma, there is less difference between the dark colours than the
light ones. In common with other models which have ‘leaves’ of constant hue,
such as Munsell and NCS, low chroma colours near the achromatic axis are
closer together than vivid, high chroma colours. CIE standard illuminant D65
was used to select the samples, and their CIE tristimulus values are published.

4.5.4 Coloroid system

This system is used primarily by architects and interior designers in Europe. As
with the DIN system, the measure of hue is related to dominant wavelength. The
spacing between hues was determined from a series of subjective assessments.
Chroma is measured on a scale from 0, for the D65 reference white, to 100 for the
spectrally pure colours. Lightness is related to CIE luminance, Y, with a square
root weighting to give a perceptually even scale. Formulae have been published
to convert between Coloroid specifications and CIE xyY values.

University of Manchester 45
 4.5.5 OSA cubo-octahedron
The Optical Society of America have designed a colour space which avoids the
uneven spacing of hue wheel systems. It is based on a rhombohedral crystal lat-
tice structure, where each point has twelve closest neighbours in 3D. The unit
solid which stacks in this fashion is a cubo-octahedron (a cube with all eight cor-
ners sliced off). This space lattice provides an even sampling of the colour space,
and allows samples to be readily ordered in planes which are not parallel to the
axes. This is intended to help designers see new scales and arrangements of col-
our.

The three parameters in the OSA model are lightness (L), yellowness (j, French
jaune) and greenness (g). The selection of colours for the lattice points was the
result of over thirty years of colour matching experiments. D65 is used as an
illuminant and, unusually, the CIE 1964 supplementary observer is used. This
poses problems for the use of OSA in a computer graphics system, because moni-
tor chromaticities are measured for the 1931 standard observer.

4.5.6 TekHVC
This relatively new system is a development of the polar coordinate L*C*Huv
form of CIELUV. It has three parameters:

H = Huv - offset

V = L*

C = 7.50725 C*

where offset is the angle in the CIE UCS between the u’ axis and the line joining
the selected white point to a particular ‘best’ red. This means that a hue angle of
0° always points to this red regardless of the white point. However, a hue angle
of 90° will still point at different colours as the white point is changed. The scal-
ing factor is intended to make the visual effect of a change in C the same as that
of a similar change in V or H.

4.5.7 Colour naming scheme

This non-geometric model produces an English name for a colour. There is an
achromatic axis with seven points along it: black, very dark gray, dark gray,
gray, light gray, very light gray and white. Six hues are named: red, orange, yel-
low, green, blue and purple. There are three intermediate points between each
hue, for example between yellow and green are, in order: yellowish green, yellow
green, greenish yellow. Hues are modified according to saturation (greyish, mod-
erate, strong or vivid) and lightness (very light, light, medium, dark, very dark).
So an example of a colour specified using this scheme would be ‘greyish light
orangish red’.

There are problems implementing this model in a computer graphics system in

that the colours are specified as Munsell colours, which must be converted to CIE
XYZ with a look up table. Also, it can describe very few colours (about 340).

46 Computer Graphics and Visualisation

4.5.8 X11 named colours
This is another naming scheme, although not as regular as CNS. Intended for
the specification of such items as window borders and mouse cursors in the X
Window System, it is essentially an enumeration scheme, with colours described
as ‘Dodger Blue’ or ‘Wheat’ for example. The specification of each colour is given
in terms of the RGB colour model. This has the inherent problem that colours
look different on different monitors, so that they do not match people’s idea of the
names. A number of ‘alternative’ specifications have been proposed, each of
which looks correct on the particular monitor of the person who designed it.

The X11 named colour list has largely been superseded, from X11R5 onwards, by
the much more comprehensive colour management services (Xcms). This allows
the specification of colour as RGB, HLS, CIEXYZ, CIExyY, CIELUV, CIELAB,
CIEu’v’Y or TekHVC and provides interconversion routines between these. The
chromaticity coordinates of the monitor phosphors and white point have become
a property of the root window, inherited by all other windows.

4.5.9 Pantone
This is a proprietary system for specifying colour, widely used in the commercial
graphic design world. Originally it specified a large series of pigments for spot
colour. This has now been extended to the Pantone process colour range
which relates colours to percentages of standardised, Pantone certified cyan, ma-
genta, yellow and black process inks used with standardised screen angles. The
process colours can vary markedly from the ‘equivalent’ spot colour.

Pantone is generally used with a book of printed samples; some applications

have a Pantone license and can produce on-screen colour, but this is only a rough
approximation as the colours are specified in the RGB colour model. Pantone is a
method of getting precise results, particularly for spot colour, but is tedious to
use. CIE tristimulus values for the samples are available. Other similar systems
include Focoltone and TrueMatch.

4.5.10 SML space

This tristimulus space is similar to CIE XYZ except that human cone pigment
spectra are used as matching functions. Colours are thus defined in terms of the
degree of excitation produced in each of the three cone types, before any visual
processing is performed by the nervous system. SML space has been used in vi-
sion research, particularly in the design of colour blindness tests and in
visualising the effects of vision defects. Given the specification of the particular
cone pigment spectra used, it is possible to interconvert between SML and CIE
1931 XYZ.

4.5.11 Hunt-ACAM model

Developed to predict the appearance of colours irrespective of viewing conditions,
this complex model requires two sets of measurements with different light
sources. Using the CIE XYZ model as a base, it tries to take into account factors

University of Manchester 47
 such as adaptation, chromatic induction, and simultaneous contrast. The
Hunt-ACAM model gives numerical predictions for colourfullness, saturation, in-
tensity, lightness, and brightness.

48 Computer Graphics and Visualisation

5 Colour output

5.1 Displaying colour

Interactive computer graphics are displayed on a colour monitor. The signals to
drive the monitor are generated by the video circuitry of the graphics worksta-
tion. A broad understanding of these components helps explain the limitations
which are met with when displaying colour, and how to minimise or work around
these restrictions.

5.1.1 Colour monitors

The majority of colour monitors in use today for computer graphics use a cath-
ode ray tube (CRT), similar to that found in a television, to generate the
picture. Other technologies, such as active matrix colour liquid crystal displays
(LCD), do not at present give the high quality colours needed for computer
graphics although they are widely used for less critical applications such as port-
able computers.

The principle of a CRT is that one or more electron guns produce variable
amounts of electrons in response to an applied voltage. The electrons are acceler-
ated towards the front of the tube by applying a large positive voltage to a grid.

Electron Focussing Deflection

guns plates coils

Accelerating Shadow
grids mask

Figure 40: Anatomy of a CRT

The front of a colour tube is covered with three types of phosphor, which emit
red, green and blue light when hit by electrons. Monochrome and greyscale moni-
tors have only a single colour of phosphor. Electron beams from the guns are
swept from top to bottom and left to right by the deflection plates to cover the

University of Manchester 49
 screen area, and the voltages applied to the three guns are varied to adjust the
intensity of the electron beam and hence the brightness of light emitted. A
shadow mask is used to ensure that the electron beam from each gun can only
fall on the appropriate type of phosphor.

5.1.2 Monitor gamut

Producing different colours by variable mixture of light from three coloured phos-
phors is very similar to the colour matching experiments described in section 3.1
except that:

• the light from the phosphors is not as saturated as a pure spectral colour
• negative values cannot be applied to the guns.

These differences mean that some visible colours cannot be reproduced on a CRT.
The range of displayable colour is termed the gamut and varies for different
makes and models of monitor. It may conveniently be depicted on a CIE 1976
UCS diagram, where it forms a triangle bounded by the monitor primaries. Each
secondary lies on the line connecting the appropriate primaries, because the col-
ours are additive. The white point should correspond to equal maximal output
from the three guns. The example below (Figure 41) shows the gamut of the
monitor on a VAXstation 3540 workstation.

0.6 525 550

575
600
500 675-780

0.4

v’

475
0.2

VAXstation 3540

monitor whitepoint
450
400
0.0

0.0 0.2 0.4 0.6

u’

Figure 41: The VAXstation 3540 gamut on the CIE 1976 UCS diagram

50 Computer Graphics and Visualisation

The choice of monitor primaries is a trade off between obtaining a large gamut
and making the display sufficiently bright. As the ISO luminous efficiency func-
tion (Figure 7) shows, the extremes of the visible wavelengths are seen as very
dim. So the primary in the long wavelength corner tends to be a bright, orangish
red rather than a dim deep red; similarly the primary in the short wavelength
corner tends to be a fairly bright blue rather than a very dim violet.

The gamut of a monitor shrinks as the ambient light level increases, a fact which
will be familiar to anyone who has tried to use a monitor in bright sunlight. Am-
bient light is reflected back from the monitor, adding white to all colours. This
means that black becomes a dark grey. All colours move towards the white point,
the darkest colours moving most. So, as the ambient light level is increased, typi-
cally deep blues are lost first, and only the lightest colours such as yellow and
white can still be seen at high ambient light levels.

Appendix B gives details of how to calibrate a colour monitor for use with CIE
colours.

5.1.3 Factors affecting monitor quality

The colour fidelity and ergonomics of a colour monitor can be adversely affected
by a number of factors:

• misconvergence: the electron beam does not hit the correct pixel. This results
in blurring of the edges of shapes and upsets the colour balance; if for exam-
ple the green gun is also lighting up the red pixels to an extent, then all
greens will be tinged with yellow (the secondary colour resulting from a mix-
ture of green and red). The gamut will clearly be reduced, the position of the
green corner of the gamut triangle moving towards the red corner in this ex-
ample. Misconvergence tends to be most apparent at the edges of the display
and in older monitors. Solution: Many monitors have internal controls to ad-
just convergence. Have these adjusted by a competent service engineer. Use
a degauss button regularly, if there is one. Do not site monitors next to mag-
netic fields, such as loudspeakers or power cable conduits.
• flicker: caused by the refresh rate of the screen being too low, or the use of
an interlaced display (where the electron beam traces all the even lines, then
all the odd lines). Solution: do not use a video mode of higher resolution than
the monitor can cope with. Do not use interlaced modes.
• phosphor ageing: over a period of a year or so, the brightness of the phos-
phors will fall. Blue is affected faster than red or green. Solution: Do not rely
manufacturers data for old monitors; have the values measured. For accurate
work, use an auto-calibrating monitor.
• gun interaction: the intensity of the electron beam depends on the power be-
ing produced by the other two guns at the time. Also, the intensity of a white
pixel will be different if the rest of the screen is all white or all black, because
of power drain. Solution: avoid cheap monitors with inadequate power sup-
plies.

University of Manchester 51
 5.1.4 Video circuitry
The image displayed on a computer graphics monitor is composed of a two di-
mensional array of dots, termed pixels. These are the smallest addressable
areas on the screen whose colour can be individually changed. The video image is
defined by an area of memory in the computer, the video RAM (Random Access
Memory), which in most workstations can be written to at the same time as the
video circuitry is reading from it. Graphics workstations typically write to this
memory with a mixture of both software and specialised hardware which per-
forms common tasks (such as drawing polygons). Video RAM is read
continuously by the video circuitry, which scans each pixel in turn and sends the
values as a serial stream to be converted into monitor signals.

Considering the video RAM to be a two dimensional array, displays differ in both
the size of this array and the colour resolution (number of bits per pixel). To-
gether with the physical size of the monitor, this defines the spatial resolution
(in pixels per inch) and the total number of simultaneously displayable colours.

Monochrome devices use one bit per pixel, so each pixel can be on or off, white or
black. Greyscale devices use more bits per pixel, the total number of displayable
greys being 2n, where n is the number of bits per pixel, typically 8. The binary
number stored in video RAM for each pixel in turn is accessed by the video hard-
ware and converted to an analogue voltage using a fast digital to analogue
converter (DAC) as shown in Figure 42. This voltage is used to modulate the in-
tensity of the electron beam in the monitor and so give different brightnesses.

DAC Gun
Video RAM Screen
Figure 42: A greyscale display.

Colour devices used in computer graphics typically use 24 bits to represent each
pixel. These are organised as three groups of eight, giving 28 = 256 levels of in-
tensity for each of the red, green and blue guns; 16.7 million colours in all. There
are thus three DACs. This, plus the cost of the extra memory and the colour
monitor, is why colour displays are more expensive than monochrome or
greyscale displays. The layout of a 24 bit colour display is shown in Figure 43.

52 Computer Graphics and Visualisation

 Video RAM DACs Guns Screen

Figure 43: A 24 bit colour display.

Some displays, used for less demanding computer graphics applications, use only
eight bits to represent each pixel. Very few devices organise this into three
groups like the 24 bit displays; this would give far too few colours in most appli-
cations. Instead, each location in video RAM stores an 8 bit value which is used
to index into a table of 256 colours. These colours are specified to a greater preci-
sion than 8 bits; 18 or 24 is common. The total range of colours is termed the
palette; the table of selections from this palette is called the colour look-up ta-
ble, or CLUT. Single chips containing a CLUT and three DACs are available, the
combination being referred to as a RAMDAC.

Although the total number of colours in the palette can be as high as the total
number of colours in a 24 bit display, only 256 of them can be used in any one
image. This configuration is an example of indexed colour, whereas the 24 bit
display described previously is an example of a direct colour system. The layout
of an 8 bit indexed display is shown in Figure 44.

CLUT

0
1
...

255

Video RAM DACs Guns Screen

Figure 44: An 8 bit indexed display.

University of Manchester 53
 It is significantly faster to rewrite the data in the colour lookup table (256 en-
tries, 3 × 8 bits, so 768 bytes) than to change the colour of each pixel in video
RAM (typically 1280 × 1024 entries, 8 bits, so around 1.3 million bytes). Rewrit-
ing the CLUT can be used to provide fast animation of an image with few
colours.

A refinement of the 24 bit display uses this indexing technique for each 8 bit
group, to index into a table of (typically 12 bit) colour values. There are thus
three of these tables, one for each gun, and three 12 bit DACs are used. Whereas
the entries in the CLUT of an 8 bit indexed display are independent of one an-
other, the entries in this system are typically ordered to form a colour scale. This
allows the maximum value and response curve of each DAC to be changed, to
compensate for drift or ageing in the calibration of the monitor.

5.1.5 Gamma correction

One of the assumptions made when converting between XYZ and RGB is that
colours are linearly additive. This assumption is invalid for a number of reasons,
primarily because linear increases in the voltage applied to the guns does not
produce a linear increase in luminance. The light produced by a phosphor is pro-
portional to the electron beam power, rather than the gun voltage.

Power = voltage × current

Current ∝ grid voltage1·5

So Luminance ∝ voltage2·5

In practice, luminance is proportional to the DAC voltageγ, where γ is in the

range 1·5 to 3·0. Thus, the values in video RAM or in the CLUT should be ad-
justed to compensate for this. Some display hardware has this correction built in.
Because the spacing of values has the same minimum (0) and maximum (255)
values, but is non linear, one result of gamma correcting the values in video
RAM is a decrease in the number of available colours. This is why some systems
use 24 bits to represent each pixel, but then use three 12 bit lookup tables to per-
form gamma correction and drive the DACs, maintaining the full range of
colours. An example of gamma correction is given in Appendix A.

5.1.6 Viewing with non-RGB models

Some display software allows colour to be specified in a standard, device-
independent format such as one of the CIE models. For example, PHIGS, GKS
9x, and ODA require the display to provide this facility. The current version
(X11R5) of the X Window System gives a means to implement this as it accepts
colour specifications in CIE XYZ, CIELUV, CIELAB and TekHVC among others;
mapping to RGB is handled transparently. This allows applications to specify the
same colour on a variety of monitors.

A problem with using a non-RGB model is that unprintable, outside-gamut col-

ours can be specified. These must be dealt with in a device dependent way. A
colour management system can be used for this. The problem of gamut map-
ping and the treatment of out of gamut colours is discussed in Section 5.7

54 Computer Graphics and Visualisation

5.2 Colour video
Video equipment has some similarities to computer graphics monitors, in that
the display is an array of pixels on an RGB monitor. There are differences, how-
ever, to do with the mechanics of encoding the signal and the different working
practices in that sector of the industry. Familiar sounding terms may have new,
different meanings in a video context.

Because animated sequences of computer graphics are frequently stored as a

video, it is helpful to have some understanding of video technology as it applies
to the use of colour. Computer graphics may then be designed with the inherent
limitations of the medium in mind. It should be remembered that the very best
professional video equipment will still give much worse resolution and colour fi-
delity than even a modest graphics workstation.

5.2.1 Broadcast monitors

Colours are specified in the RGB space of standardised broadcast monitors,
whose phosphor chromaticities, white point, and gamma value are tightly speci-
fied by the appropriate standards body. This enables an engineer to check the
quality reliably in any studio. It also enables a computer graphic to be con-
structed to those standards, so that it will look correct when viewed. Table 3
shows the definitions for three popular standards.

Red x,y Green x,y Blue x,y

NTSC 0·670 0·330 0·210 0·710 0·140 0·080

SMPTE 0·630 0·340 0·310 0·595 0·155 0·070

EBU 0·640 0·330 0·290 0·060 0·150 0·06

Table 3: Chromaticities of standard broadcast monitors

Broadcast standards in the United States of America are defined by the National
Television Systems Committee (NTSC). The NTSC standard is also used in other
countries, such as Japan. It specifies illuminant C for the white point and a
gamma value of 2·2.

The United Kingdom and Europe use standards laid down by the European
Broadcasting Union (EBU). These use a D65 whitepoint and a gamma of 2·2.

Colour correction in modern studios is done on a studio monitor which conforms

to the code of practice of the Society of Motion Picture and Television Engineers
(SMPTE).

The gamuts of these three monitors are shown in Figure 45. To provide the cor-
rect colour in high quality computer graphics, the CIE XYZ values should be
converted to RGB using the SMPTE monitor chromaticities.

University of Manchester 55
 0.6 525 550
575
600
500 675-780

0.4

v’
D65 whitepoint
475
0.2 Illuminant C

SMPTE monitor

EBU monitor
450
NTSC monitor
400
0.0

0.0 0.2 0.4 0.6

u’

Figure 45: SMPTE, EBU and NTSC monitor gamuts

5.2.2 Gamma correction

In contrast to general computer graphics practice, where gamma correction is the
last step before display, the RGB signals which are to be recorded on video are
gamma corrected before encoding. The alternative would require gamma correc-
tion circuitry in every replay monitor and, if the signal were to be broadcast, in
every TV set. It is easier and more cost effective to produce a single stable,
closely tracking corrector in the studio.

Because of this, it is important that computer graphics images which are to be

recorded on video should first be gamma corrected. Gamma correction in a com-
puter graphics context is discussed in Appendix A. While any arbitrary
transform can be applied to a computer graphic image, the standards for video
assume that the input signals are coming from a TV camera and that the gamma
correction will be performed by an electronic circuit. The graph of a pure power
law meets the origin vertically, which corresponds to infinite gain in a circuit.
This is clearly impossible to attain, and would give severe problems with noise
amplification, so gamma correction in a video context uses a modified power law
with a linear section near the origin. To preserve compatibility, the same trans-
formation should be used for computer graphics that are to be stored on video.

The formulae given below are defined in CCIR Rec. 709, the international stan-
dard for high definition television (HDTV). Older video standards have less
precise definitions for gamma correction, ignoring the infinite gain problem.

56 Computer Graphics and Visualisation

CCIR Rec. 709 formalises the range of similar values used in modern studio
equipment, and these equations are suitable for use in computer graphics.

For values in the range 0.018 to 1.0:

Rγ = 1·099 R 0.45 - 0·099

Gγ = 1·099 G 0.45 - 0·099

Bγ = 1·099 B 0.45 - 0·099

For values in the range 0.0 to 0.017, the slope of the curve is limited to 4.5:

Rγ = 4·5 R

Gγ = 4·5 G

Bγ = 4·5 B

5.2.3 Bandwidth
In computer graphics, a picture is thought of as a 2D array of pixels, each of
which can be any colour. This is simple and intuitive, and the very high speed
electronics which make this possible go unremarked most of the time. In video
recording, a picture is thought of as a waveform which shows how the intensity
or colour of the picture varies as the electron beam scans across the display. As
the lines on screen are traced sequentially, a picture in video terms is thought of
as a 1D waveform.

A slowly changing picture corresponds to a low frequency waveform. The greater

the changes between one pixel and the next, the higher the frequency required to
encode it and thus the greater the total bandwidth. This is the fundamental dif-
ference between computer graphics and video; adjacent pixels in each horizontal
scanline are related. There is however no special relationship between a pixel
and the one above or below it. They are on different scanlines.

A typical computer graphics image, where adjacent pixels could be completely

different, requires an extremely high bandwidth to represent it. In video record-
ing, the available bandwidth is severely limited and so the amount by which one
pixel can vary from the one that preceded it is limited. The maximum number of
pixels in a scanline in 768, but the effective horizontal resolution will thus be
less.

5.2.4 Refresh rate

A typical computer graphics monitor may have a vertical resolution of 1024 lines
and be refreshed 72 times a second or more. This means that 1024 × 72 = 73,728
scanlines are refreshed each second. A video recorder has 625 lines, of which 50
are used for other purposes, and is refreshed 25 times a second; in other words it
can only draw (625 - 50) × 25 = 14,375 scanlines a second.

University of Manchester 57
 To reduce flicker, the even scan lines are drawn first, followed by the odd scan
lines. This raises the effective screen refresh rate to 50 times a second, provided
there is some correlation between one scanline and the next. When the picture
comes from a camera this is often the case, but is less likely in a computer graph-
ics image. Horizontal lines which are one pixel wide will flicker noticeably. The
effective horizontal resolution is thus reduced from 575 to between 200 and 300.

5.2.5 Video luminance

Colours are converted from RGB to a video luminance signal and two colour
difference signals. This is similar to the encoding of three colour signals to lu-
minance and colour difference signals in the eye, which was described in Section
2.6, and is done for the same reason: to make the best use of the available infor-
mation carrying capacity.

As originally calculated in the 1950s, the luminance encoding corresponded to a

weighted sum of the luminance contribution from each phosphor, using the moni-
tor chromaticities of the original NTSC broadcast monitor.

This gives the formula for video luminance which is:

Y = 0·299 Rγ + 0·587 Gγ + 0·114 Bγ

This has now become standardised (CCIR Rec. 601-1) and the equivalent decoder
is built into all video recorders and TV sets.

Over the years, the phosphors used in colour televisions have changed, so that
modern sets are quite different from the NTSC primaries. To provide more useful
colour monitoring, the Society of Motion Picture and Television Engineers
(SMPTE) published a standard code of practice which defined new chromaticites,
shown in Table 3. Colour balancing is now done using the SMPTE broadcast
monitor. Because the NTSC monitor is not representative of modern broadcast
monitors or TV sets, the video‘ luminance’ signal is no longer the correct formula
to calculate the luminance which would be measured from the monitor. Apart
from making a less than optimal use of the information capacity of the system,
this does not matter. It is more important that the encoder and decoder work cor-
rectly as a pair.

Use the above formula to calculate the video luminance signal which will be re-
corded onto tape. This is the same regardless of the monitor which is being used
for colour balancing.

To calculate the actual measured luminance, for example to convert a full colour
image to greyscale, calculate directly from the chromaticities of the monitor in
question using the procedure in Appendix B. For example, for an SMPTE moni-
tor the formula would be:

Y = 0·2122 R + 0·7013 G + 0·0865 B

Notice that the RGB values are should not be gamma corrected in this formula.
It is unfortunate that the symbol Y has these two related but distinct meanings.
The chromaticity of the video luminance signal Y is at the system white point

58 Computer Graphics and Visualisation

(usually D65) whereas the CIE Y primary is an imaginary colour with
chromaticity u´= 0, v´ = 0·6.

As an additional trap for the unwary, HDTV will use a different formula for
video luminance. Being a new standard (SMPTE 240M) for a much improved
video and broadcast system, it does not need to maintain compatibility with an
installed base of older decoders. So the video luminance encoding was calculated
from the SMPTE monitor set:

Y = 0·2122 Rγ + 0·7013 Gγ + 0·0865 Bγ

5.2.6 Video chrominance

Having produced a video luminance signal, another two channels are required to
carry enough information to reconstruct the original RGB values. These are
called chrominance, as they were originally related to the chromaticity differ-
ence between the colour being encoded and the white point. Because video
luminance is no longer equal to CIE luminance, this is no longer the case.

The chrominance signals are produced by subtracting the video luminance from
any two of the three (RGB) channels. Green makes the most contribution to the
video luminance, so it is the red and blue channels that are used to make
chrominance signals.

C1 = Rγ - Y

C 2 = Bγ - Y

Because of these formulae, chrominance is also sometimes called colour differ-

ence.

Clearly, the chrominance signals can be positive or negative. Putting the RGB
values of the primary (red, green, blue) and secondary (cyan, magenta, yellow)
colours into these equations shows that C1 varies between ± 0·866 and C2 be-
tween ± 0·701.

Converting an RGB signal to a luminance (Y) and two chrominance (C1C2) sig-
nals is a linear, reversible process and is the starting point for all video encoding.
Of itself, it produces no limitations or distortions of colour quality, although later
stages certainly do.

5.2.7 Analogue component video

These video systems record the three components of the signal - video luminance
and two video chrominances - separately, so that on playback the picture looks
very similar to the original RGB signal. The main limitations are the signal to
noise ratio, stability and linearity of the analogue recording system. More costly
professional broadcast equipment gives a better picture than industrial grade
equipment.

University of Manchester 59
 This type of system is used for professional video equipment. Examples are the
Sony Betacam system, widely used for electronic news gathering and studio qual-
ity mastering, and the Panasonic MII system.

In analogue component recording, the chrominance components are scaled so

that the somewhat awkward minimum and maximum values become ± 0·5:

Pb = 0·5 / (1 - 0·114) C1

Pr = 0·5 / (1 - 0·299) C2

Additionally, the Pb and Pr components are subsampled by a factor of two to

give some signal compression. This bandwidth limitation is similar to replacing
pairs of pixels with their average colours. The visual effect of this is less than it
would seem, however, as the video luminance is left intact. Subsampling is only
done on the chrominance components. The human eye is poor at identifying ex-
tremely small areas of colour, and can only see the difference in brightness. To
the extent that video luminance represents the brightness of the video signal,
this optimisation is successful. The result is a very slight smearing of small de-
tails in the image.

5.2.8 Digital component video

Like analogue component video, the three components are stored separately and
the chrominance components are subsampled by a factor of two. The resolution is
fixed by the number of bits used to code each component. There are two methods
defined, either 8 bits or 10 bits per component; at present, only the 8 bit form is
in general use.

Video luminance is recorded as an 8 bit unsigned number and the chrominance

components as 8 bit signed numbers. Not all 256 codes are used, however; the
standards for digital component video, CCIR Rec. 601-1 and CCIR Rec. 656 spec-
ify that some codes at the top and bottom of the range are left unused for
headroom and footroom.

Y = 219 Y + 16

Cr = 224 C1 + 128

Cb = 224 C2 + 128

The big advantage of digital component video is the lack of noise and the possi-
bility of many generations of copying without degeneration of the quality. Digital
component video recording equipment is however considerably more expensive
than analog component equipment.

If computer graphics are being made specifically for recording to digital compo-
nent video, the values of R G and B should be gamma corrected as floating point
values then converted directly into the range 16 to 226 rather than the usual 0 to
255. This avoids introducing two sets of round off errors.

The three components may then be calculated directly:

60 Computer Graphics and Visualisation

 Y = (77 Rγ + 150 Gγ + 29 Bγ) / 256

Cb = (131 Rγ - 110 Gγ - 21 Bγ + 128) / 256

Cr = (44 Rγ - 87 Gγ + 131 Bγ + 128) / 256

Other systems which use YCbCr or derivatives of it include the JFIF image for-
mat from JPEG, and the PhotoCD format from Kodak.

5.3 Broadcasting colour

The broadcast system used in Western Europe, PAL, is a development of the
older American NTSC system. The same considerations given in the video sec-
tion also apply to encoding for broadcast. However there are additional
limitations which must be taken into account if acceptable quality pictures are to
be produced.

Video luminance is combined with a composite chroma signal (which has been
modulated onto a colour subcarrier), colour burst and synchronisation signals to
form a single composite video signal ready for broadcast. The details of this proc-
ess need not concern us here, but once video luminance and chrominance have
been mixed together they can never be fully separated again, even by the best
quality decoders. There is inevitably some interference between luminance and
chrominance channels. This gives rise to many types of colour artefacts.

Composite video recorders record a broadcast signal (or some intermediate stage)
and are thus subject to the same limitations as if the signal were to be broadcast.

5.3.1 UV encoding
Formation of U and V components is the first step in preparing a video signal for
broadcast. The chrominance signals are scaled so that the sum of the video lumi-
nance and the composite chrominance is less than 1.34:

U = 0·493 C2

V = 0·877 C1

This is done so that the final composite signal stays within the safe limits of car-
rier modulation. The U and V components are then subsampled by a factor of two
to reduce bandwidth.

Some computer graphics systems designed specifically for broadcast use allow
colour specification directly in YUV. While not particularly easy to use, this does
have the advantage that untransmissible colours are not inadvertently produced.

5.3.2 Composite chrominance

Composite chrominance is produced by modulating the U and V signals onto a
4.43 MHz colour subcarrier, t :

University of Manchester 61
 C = U cos(t) + V sin (t)

Figure 46 shows how the phase and amplitude of this signal correspond to the
original U and V components.

V
A
φ
U

Figure 46: Phase angle (φ) and amplitude (A) of composite chroma.

5.3.3 S Video
S video is a compromise between full component recording and composite record-
ing. Video luminance and composite chroma are recorded as two separate
signals; while there is a degradation in colour quality caused by the use of the
colour subcarrier, there is no interference between video luminance and compos-
ite chrominance.

The most common example of an S video system is S-VHS. The input connectors
are often labelled Y/C, to show that video luminance (Y) and composite
chrominance (C) are kept separate. Other formats which record separate Y and C
signals are U-Matic and Hi-8 video.

5.3.4 PAL broadcast

The composite chrominance signal is combined with the video luminance signal,
synchronisation and other control information to form a single, composite signal.

As a result of mixing the luminance and chrominance signals, some very bright,
saturated RGB colours map to signals which cannot be broadcast; either because
they overload the transmitting equipment or because they give shimmering, un-
stable colours. Although such highly saturated colours are not commonly found
in film of the real world, they are a problem in computer generated images, and
the colours must be mapped to broadcastable ones by reducing the luminance or
saturation.

The constraints on acceptable colours are firstly that the amplitude of the com-
posite chroma must be less than 53, and secondly that the sum of video
luminance and composite chroma must be less than 1·2.

62 Computer Graphics and Visualisation

 Broadcast signals are sensitive to phase errors in transmission, which alter the
chrominance phase f and thus change the hue. To correct this for small errors,
the PAL system inverts the Phase of V on Alternate scan Lines, hence the name.
By averaging consecutive scanlines, the phase error is cancelled at the expense of
a further reduction in vertical resolution.

Another consequence of mixing chrominance and video luminance is that the two
signals interact. High frequency video luminance information is decoded as
chrominance, giving shimmering rainbows of colour on detail such as finely
spaced lines. The colour subcarrier is incompletely removed from the video lumi-
nance information, giving a periodic variation in brightness across a scanline.
Rapid transitions in colour produce a flickering edge, an effect that can be re-
duced by smoothing the image before it is encoded.

Domestic VHS video records a composite video signal, and thus suffers from all
the defects in colour quality that have been noted for PAL broadcast. This should
be borne in mind when planning a video animation which is to be shown at a con-
ference or presentation. The most common format to use is VHS; so although the
material is never broadcast, it is subject to all the limitations of broadcast media.

5.3.5 NTSC broadcast

This is the broadcast system used in the USA and Japan. As it was the basis for
PAL, most of the comments in the previous section also apply to NTSC. Some dif-
ferences from PAL:

• 525 lines rather than 625 lines vertical resolution

• screen refresh is 30 rather than 25 times a second
• no phase alternation, so there are hue errors
• there is even less bandwidth available than in Europe

As originally defined, NTSC used I and Q rather than U and V, defined by:

I = V cos(33°) – U sin(33°)

Q = V sin(33°) + U cos(33°)

This is a simple rotation and flip of the axes so that Q contains blues and violets;
as was seen in Section 2.7, these are the colours for which the spatial resolution
of the eye is poorest. To cope with a narrow available bandwidth, the Q signal is
broadcast at much lower resolution than Y or I. Considering the bandwidth of
the Y signal to be 100%, I uses 25% and Q only 10%. Modern production equip-
ment uses U and V rather than I and Q; the decoder cannot tell the difference.

5.4 Coping with insufficient colours

Sometimes, a graphical image containing a lot of colours must be output to a de-
vice with less colours. For example, a 24 bit image may need to be displayed on
an 8 bit display. Or an 8 bit image may have to be output on a printer having

University of Manchester 63
 only eight colours. There are three classes of technique that can be used:
quantisation, dithering and halftoning.

5.4.1 Quantisation
Each original colour is mapped to the nearest of a subset of new colours. The no-
tion of ‘nearest’ implies that the distances between colours are measured and
compared; thus, the colours should be quantised in a colour space where the dis-
tance between points correlates with their perceived colour difference. CIE LUV
is a suitable colour space, although others can be used. Distance may also be
weighted to take advantage of perceptual effects. For example, preserving lumi-
nance may be more than preserving hue.

There are four main quantisation methods:

5.4.1.1 Uniform Quantisation

Each original colour is mapped to the nearest of a standard, evenly distributed

range of colours. This is computationaly simple, and may be the only method pos-
sible if the palette of colours is fixed. It is also of use with animated graphics;
each frame must use the same palette of colours to avoid noticeable flickering. It
gives poor results, however, when the original colours are not evenly distributed
in the colour space.

To calculate a standard colourmap, a colourspace is chosen and sampled at regu-

lar intervals. A problem with using device independent colour spaces when
resolution is limited is that some of the samples represent out of gamut colours.
It is therefore customary to sample the devices native colour space, such as RGB,
directly. This has the disadvantage that equally spaced samples are not at equal
perceptual intervals.

Sampling in RGB space should use an equal resolution for each axis. For exam-
ple, in an 8 bit system with 256 colours available, 6 levels of red, green and blue
may be used to give 216 different colours. This ensures that greys do not have
objectionable tints. Having unallocated colours may be an advantage when a
colourmap is shared between concurrent applications.

5.4.1.2 Median cut

This algorithm has two distinct phases. Firstly, the colours are divided into
groups. Secondly, each group is assigned a new colour. To group colours, the
smallest rectangular, axis-aligned space is calculated that will completely en-
close all the colours. This is shown in Figure 47a, using only two dimensions for
clarity. The rectangle is cut in two across its longest axis so that the resulting
rectangles contain equal numbers of colours (Figure 47b). The process is re-
peated, shrinking each new rectangle to its smallest volume (Figure 47c) and
cutting the larger in two (Figure 47d) until enough rectangles have been pro-
duced. To assign a new colour for the group, either the centroid of the rectangle
or the group mean can be used.

64 Computer Graphics and Visualisation

 a b

c d
Figure 47: Median cut.

5.4.1.3 Histogramming

The most frequently used original colours are retained intact. The remaining col-
ours are assigned to nearest of these popular colours. It is customary to drop
some precision when constructing the frequency histogram, so that groups of col-
ours which are close count as a single colour and are assigned the group mean.
This allows other, less popular but significantly different colours to be retained
intact. Another optimisation is to leave some room in the colour map for infre-
quent colours which do not match any of the new colours well.

Whilst it is possible to construct a histogram across the entire colour space, this
requires a lot of memory. It is more common to construct separate histograms for
each axis. For example, to quantise down to 256 colours in CIELUV space one
might histogram over 50 bins in L* and 25 bins in u* and v*. This would give 2%
steps in lightness, which are just noticeable. Of these, the commonest 10 in
L*and the commonest 5 in u* and v*would be used to construct a new colourmap
with 10 × 5 × 5 = 250 colours, leaving 6 for those outliers furthest from the new
colours.

5.4.1.4 Variance minimisation

A statistical approach, this is similar to fitting a line to a set of points by least

squares analysis. For each group of similar original colours, the variance in the
distance to the new colour is calculated. The aim is to minimise the sum of these
variances. By minimising the variance rather than the mean distance, it avoids
the situation where some groups of colours are widely separated while others
cluster tightly.

University of Manchester 65
 high variance low variance

Figure 48: Variance minimisation.

There are clearly many possible ways to partition the colour space, and it would
be impossible to test them exhaustively. Some heuristic variance based methods
therefore select a low, rather than optimal, variance. It is possible to calculate
the optimal solution by a similar method to the median cut algorithm.

A plane is swept along each of the three axes of the colour space, and at each
position on each sweep the variance on each side of the plane is calculated. The
point which minimises the variance on both sides is noted. The colour space is
then cut across whichever of the three axes gave the lowest result. This is re-
peated on the side with the largest variance until sufficient subdivisions have
been made.

5.4.2 Dithering
Dithering is most commonly encountered as a means of simulating grey scales on
a monochrome device by printing black and white. However it can also be used in
a similar manner to simulate more colours by displaying two or more colours
close together; from a distance the eye will mix these to give the effect of more
colours.

To dither an area in a given colour, the available colours that most closely resem-
ble it are selected and a percentage mixture calculated. Suppose, for example,
that an unavailable lime green is to be simulated with 70% pale green, 20%
lemon yellow, and 10% dark green. For each pixel, a random number is gener-
ated in the range 1 to 100. If it is below 70, the pixel is coloured pale green;
between 70 and 90, lemon yellow, and above 90, dark green.

Clearly this technique will give the best results when the colours to be mixed are
fairly close, and large areas are dithered. Accurate colour discrimination is in
any case less easy on small areas.

A technique can be borrowed from monochrome dithering: error diffusion. For

each pixel, the error between the intended and actual colour is calculated and
used to adjust the intended colour of pixels near to it. The amount of error is
weighted according to how far away the pixels are, so that the error is ‘diffused’.
To keep the technique simple, only those pixels which have not yet been dithered
are adjusted. This is shown in Figure 49. The shaded pixels have been dithered,

66 Computer Graphics and Visualisation

and the pixel containing the star is currently being processed. The weighting fac-
tors decrease with distance from the current pixel.

× 0.5 0·25

0.125 0.25 0.5 0.25 0.125

0.063 0.125 0.25 0.125 0.063

Figure 49: Error diffusion

Because errors have both sign and magnitude, there is no net error propagated
across the image. In monochrome dithering, the error is a single number whereas
colour dithering produces three errors, one for each axis of the colour space.
There are a number of weighting schemes for diffusing errors. Figure 49 uses
Stucki weighting, for example.

Other enhancements include adding a noise function to the error, and processing
alternate lines in opposite directions; both reduce patterning.

5.4.3 Halftones
Commercial printing presses are unable to mix inks in different proportions to
achieve different colours. Instead, they lay down a grid pattern of dots for each
ink; the size of each dot is varied and from a distance, the colours appear to mix.
The dotted appearance of newspaper photographs is a familiar example of this,
using a single colour. The difference between halftoning and dithering is that the
latter keeps the same spatial resolution, whereas halftoning trades significant
amounts of spatial resolution for an increase in the number of apparent colours.

While halftoning may in principle be applied to any device, in practice it gives

best results on a device with high spatial resolution but poor colour resolution
and is thus primarily encountered in the context of four colour process print-
ing using cyan, magenta, yellow and black inks.

5.4.3.1 Screening

The process of converting a continuous tone image to a halftoned image is called

screening. This is because, in the original optical procedure, a perforated metal
screen was placed on or slightly away from a photographic film and the image
re-photographed. This process is now generally performed by a computer.

University of Manchester 67
 If the grid for each colour was aligned, each ink would be obscured by the next
one to be printed as most inks are opaque. The grids are therefore rotated rela-
tive to one another. Some angles give pronounced Moiré interference patterns. It
has become conventional to use rotations of 0° for yellow, 15° for cyan, 45° for
black and 75° for magenta to avoid this, as shown in Figure 50 and in Plate 37.
These angles give small rosette patterns. Other angles are also used, often in
conjunction with different screen frequencies for each colour.

Yellow 0 degrees Cyan 15 degrees

Black 45 degrees

Magenta 75 degrees

Figure 50: Conventional halftone screen angles.

A screened halftone is characterised by the screen angles and the frequency or

number of lines per inch (lpi). A frequency of 133 lpi is common for general print-
ing purposes. High quality book illustrations use a higher frequency such as 175;
colour daily newspapers have frequencies of around 85lpi and some colour laser
printers manage only 60 lpi.

5.4.3.2 Matrixing

Commonly referred to as matrixed dither, this is more correctly a form of

halftoning. In this method, the physical pixels are grouped into clusters of, for
example, 2×2 or 3×3 pixels. Each pixel in the cluster may be assigned a different
colour from the ones available. From a distance, the eye will perceive this as an
additively mixed colour. This works best when the available spatial resolution is
high, and the colours to be mixed are not too dissimilar. If a large matrix is used,
more colours can be produced but the image becomes visibly coarse as the effec-
tive pixel size is increased. An objectionable textured effect is a problem of the
method, caused by a repeating matrix pattern; the eye is particularly sensitive to
such regular patterns and will enhance them. Texturing can be reduced by ran-
domly selecting between a variety of different patterns. For example, in a 2×2
matrix containing three pixels in yellow and one in red, there are four possible
positions for the red pixel as shown in Figure 51.

68 Computer Graphics and Visualisation

 Figure 51: Equivalent matrixing patterns.

5.5 Printing in colour

5.5.1 Terminology
Printing may mean one of two things:

1. sending a file to a physical device, such as a laser printer, waxjet, or dye

sublimation printer, which produces hardcopy.

2. sending some data, such as a file or a photograph, to a printing works to

have halftone printing plates made. Large numbers of prints can be made
from these plates, often by a process called offset lithography.

These two procedures, although different in details, have in common that a col-
oured print is produced by mixing inks, wax, or other dyes. They can mostly be
considered equivalent, except where noted.

Printing technologies can be divided into two types. In the first, more common
type the amount of ink deposited on a particular spot has two values - some and
none. With three colours of inks, this means that only eight colours are available.
Halftoning, discussed in the previous section, is used to increase this paltry
range. Offset lithography and most laser printers, waxjets and inkjets are exam-
ples of this type. In the second type the quantity of ink deposited can be varied,
allowing a wide range of colours to be produced without any halftoning. They are
thus termed continuous tone printers, and are capable of near photographic qual-
ity. Dye sublimation printers are an example of this type.

5.5.2 Mixing inks

Printing with coloured inks is a subtractive process, in contrast to the additive
process when lights are mixed. Inks are reflective colours, and subtract from the
illuminant to produce colour as shown in Figure 52 and in Plate 27.

University of Manchester 69
 Surface gloss
Illuminant Emitted colour
reflected colour

Light source

Figure 52: Additive and subtractive colours.

This means that, in a three colour process system, the primary colours of inks
broadly correspond to the secondary colours of three colour additive mixing –
cyan, magenta and yellow - and that a mixture of all three primaries produces
black rather than white. It is also possible to print with spot colours in addition
to, or instead of, process colours. This may be done to produce colours outside a
process colour gamut, to ensure that a particular colour looks the same on differ-
ent printers, or to obtain special effects.

5.5.3 Printer Gamuts

Because printing uses subtractive mixing, the theoretical gamut of a printer is
defined on a chromaticity diagram by six points: three primaries and three secon-
daries. This is in contrast to the gamut of a monitor, which is defined by three
points as additivity ensures that each secondary lies on the line connecting its
constituent primaries. The shape of a typical printer gamut, in this case a
Tektronix Phaser IIIPXi, is shown in Figure 53.

70 Computer Graphics and Visualisation

 0.6 525 550
575
600
500 675-780

0.4

v’

0.2 475

450
400
0.0
0.0 0.2 0.4 0.6
u’
Figure 53: Example printer gamut

A good set of printing inks will have highly saturated colours to give a wide
gamut. The variation in coloured inks used in different printers is at least as
wide as the variation in monitor phosphors.

Although colour mixing is subtractive, the density of ink can be considered addi-
tive. Printing the full amount of cyan, magenta and yellow gives a density of
300%, and a brownish black colour. Adding black ink would give a total achiev-
able density of 400% (and a better black colour). In practice, density departs from
additivity in a number of ways, which not only alters the gamut but also greatly
complicates the colour separation process.

Because of this, the sort of gamut shown in Figure 53, though useful as a guide,
is not completely accurate. Professional printers and colour pre-press agencies
construct a more accurate gamut by measuring samples of many combinations of
inks. A test sheet containing samples with a known proportion of CMY or CMYK
is produced, and the CIE tristimulus values of each patch measured. Plate 38
shows part of such a sheet. The resulting data is used to construct a gamut and,
arranged as a 3 or 4 dimensional lookup table, is used to predict the proportions
of inks required to mix a given colour on that device.

5.5.4 Factors affecting quality

The number of variables affecting the quality of a finished print is enormous;
only a selection can be covered here. This is why specialist colour pre-press agen-

University of Manchester 71
 cies are used when high quality is desired, such as in book and magazine produc-
tion.

It was stated earlier that ink density could be considered additive. Many of these
variables are to correct for this assumption being only approximately true. Some
of the more important factors are:

• Ink adhesion. Inks are layered on top of each other, and there is a difference
in adhesion between printing on plain paper and on inked paper. This effect,
which varies depending on the order inks are deposited onto the paper, re-
duces additivity.
• Maximum density. A density of 400%, corresponding to the maximum
amount of each ink, would weaken most papers and result in buckling, tear-
ing and smearing. Because equal quantities of the three primaries will
produce a neutral gray, some or all of the common proportion of an ink mix is
removed and substituted with black. For example, the colour
C=20,M=30,Y=40 could be replaced with C=0, M=10, Y=20, K=20 or C=10,
M=20, Y=30, K=10. This is called grey component replacement (GCR).
Plate 39 shows 70% GCR.
• Gray balance. In practice, equal amounts of each primary do not produce a
pure achromatic grey. The proportions must be adjusted slightly to remove
the colour tints which would result, and to which the eye is particularly sen-
sitive. This problem, shown in Plate 40 is lessened by grey component
replacement.
• Under colour removal. Black ink is used to boost the ink density at high ink
levels to correct for additivity failure. This enhances contrast, and is per-
formed in addition to grey component replacement.
Measured density

C+M+Y+K

C+M+Y

Required density

Figure 54: Under colour removal

• Press registration. This applies both to printing plates and to multiple passes
of paper inside a printer. Mis-registration alters the colour of an ink mixture
from the measured, calibrated value and can introduce colour tints into
greys. Very small amounts of mis registration can affect the final result, be-
cause the degree of overlap of halftone dots and the effective screen angles
are altered.

72 Computer Graphics and Visualisation

 • Effective dot size. Measured ink density is not completely proportional to the
specified area of dots on a halftone screen. Factors such as ink viscosity and
diffusion through the paper result in dot gain, an increase the actual dot
area. Dot gains of over 25% can occur in some cases. In addition, there is a
sudden step in density at the point where individual dots become large
enough to touch, giving the effect of white dots on a coloured background
rather than coloured dots on a white background.
• Ink variability. Batch to batch variation must be considered as, unlike phos-
phors, inks are consumables.
• The precise screen angles and frequencies, and the physical resolution of the
device, determine the number of colours that can be printed.

5.5.5 Printing using non-CMYK models

Some printers allow colour to be specified in a standard, device-independent for-
mat such as one of the CIE models. For example, the current version (Level 2) of
the PostScript page description language, which is widely used in printers and
imagesetters, provides this facility.

In PostScript, a basic framework for CIE based colour models is parameterised to

define the desired colour model. Graphics are then described using that model.
Examples of possible PostScript colour models include CIE 1931 XYZ, CIE 1976
LUV, or SMPTE RGB. Recall that RGB, in a broadcast context, relates to a par-
ticular set of phosphors.

One problem with using a non-CMYK model is that colours can be specified
which cannot be printed on a particular device. These must be dealt with in a
device dependent way. Using a graphical colour selection tool or colour manage-
ment system, such colours can be avoided. For example, Plate 41 shows such a
tool displaying the difference between the gamut of a monitor (the Tektronix
XP29P PEX terminal, solid line) and a printer (the Tektronix 4396DX, dotted
line). The selected colour is shown in both the RGB and TekHVC colour models.

The problem of gamut mapping and the treatment of out of gamut colours is
discussed in Section 5.7

5.6 Colour photography

Photography is a continuous tone process, capable of fine gradations of colour
and very fine resolution. Colour film, once developed, may be output on print,
transparency, or saved to compact disc.

In a photographic print, colour is produced by mixing of light reflected from three

thin translucent layers superimposed on paper. The gamut is defined by the
chromaticities of the dyes in these layers. Because the colour is produced by sub-
tractive mixing, the gamut is not a regular shape and can only be accurately
determined by measurement of many samples.

University of Manchester 73
 Photographing the display on a graphics monitor gives good results, particularly
if the room is darkened to exclude reflections on the glass and a tripod is used.
The gamut of a colour film is typically large, and includes much of the gamut of
many computer monitors. This is why screen photography often provides a more
pleasing and accurate result than using a colour printer. Transparency (slide)
film has wider colour gamut and better colour fidelity than print film. Magazines
and journals tend to prefer transparencies to prints for this reason.

0.6 525 550

575
600
500 675-780

0.4

v’

475
0.2
D50 whitepoint

monitor whitepoint

typical transparency
450
400 HP A1097C monitor
0.0

0.0 0.2 0.4 0.6

u’

Figure 55: Comparison of monitor and slide film gamuts.

5.7 Gamut mapping

All physical devices can only reproduce a subset of the range of visible colours.
As we have seen, this subset is device dependent. There is thus a problem dis-
playing an image whose gamut is not a subset of the display device. This is most
commonly encountered when an image is produced on one device and displayed
on a second. For example, when an object displayed on a computer graphics
monitor is to be printed. For comparison, the gamut of a typical printer (a
Tektronix Phaser IIIPXi) is shown in Figure 56 together with that of a typical
graphics monitor (a VAXstation 3540 24 bit display). Large amounts of the blue
portion of the monitor gamut lie outside the printer gamut. On the other hand,
the printer can reproduce green-cyan and magenta-red areas that cannot be dis-
played on the monitor.

74 Computer Graphics and Visualisation

 0.6 525 550
575
600
500 675-780

0.4

v’

475
0.2 monitor whitepoint

D65 whitepoint

VAXstation 3540
450
Tektronix Phaser III
400
0.0

0.0 0.2 0.4 0.6

u’

Figure 56: Comparison of printer and monitor gamuts.

There are a number of strategies for transforming colours between the two de-
vices.

One simple solution is to use a direct mapping of RGB to RGB or CMY. This has
the advantage that each unique colour in the source gamut maps to a unique col-
our in the target gamut. However, none of the colours will be correct. For
example, using the pair of gamuts in Figure 56, the monitor blue would be repre-
sented by the printer blue. In this example, the printer blue is very close to the
line joining monitor blue with monitor red. It will thus appear as purple, a mix-
ture of those two colours.

While this direct mapping may be useful when colours are merely required to be
distinguishable, the general requirement is to reproduce the original colour with
the greatest perceived (rather than measured) accuracy.

One alternative is to use the full gamut of the first device, and approximate the
out of gamut colours that result. This is shown in Figure 57a.

Another alternative, shown in Figure 57b, is to use only those colours which are
in the intersection of the two device gamuts. This requires accurate sampling
and measurement of the gamut in the case of non additive devices such as print-
ers. It also introduces problems when colours must be generated automatically,
for example in shaded images.

A compromise solution, Figure 57c, is to bound the gamut of the first device by
some regular shape which approximates the gamut of the second device. This
limits the number of out of gamut colours that are produced, while allowing
lighting calculations to choose from a defined range of colours. Such a solution
would however severely limit the range of possible colours, and still be quite com-
plicated to implement.

University of Manchester 75
 Figure 57: Gamut selection strategies.

5.7.1 Dealing with out of gamut colours

Given that a particular image contains out of gamut colours, a strategy must be
developed to deal with these. The resulting image will contain colours that are
not completely correct. The aim is to reduce the visual impact of such changes.

The phenomenon of colour constancy helps to reduce the complexity of this task.
In any mapping transformation of this nature, the least noticeable change is a
saturation shift. Greyscale shifts (lighter/darker) are also not too bad provided
there are no sudden discontinuities. Overall shift in lightness of an image are tol-
erated because of colour constancy and the adaptation of the eye to different
lighting levels; white remains white.

Hue shifts are generally objectionable, particularly if a colour moves to a colour

of a different name. For example, moving a colour which was originally blue
looks steadily further from true until the blue suddenly becomes classified as
green, or purple; the perceived jump at this point is very noticeable. This is a
psychological effect, rather than a physical one. It corresponds to the third stage
of visual processing in Figure 10. Clearly, colours near a name boundary are the
most succeptible to hue shifts.

One method is to clip all colours outside the gamut to the gamut boundaries.
This has the advantage that all colours inside the gamut are unchanged and will
appear accurate. However, as shown in Figure 58, many out of gamut colours
will map to the same colour. This plays havoc with graduated ranges of colour
such as smoothly shaded lighting effects. The discontinuities in colour relation-
ship are visually obtrusive.

76 Computer Graphics and Visualisation

 Figure 58: Boundary clipping.

Another way is to uniformly scale all colours towards the grey axis, as shown in
Figure 59, so the transformed set of colours fits within the gamut. This keeps
smooth ramps of colour intact, although the colours can loose a lot of saturation.
The hues will remain constant, however, so the image will look recognisably
similar. Non linear scaling, where colours are moved progressively more the fur-
ther they are from the grey axis, can help here.

Figure 59: Uniform scaling

A good method in practice is to scale so that most - 90 to 95% - of the colours fall
within the gamut. Then the outliers are clipped to the boundary. This avoids sub-
jecting the majority of colours to a substantial colour shift, just to accommodate a
few outliers.

Although this transformation can be done once, to map the entire monitor gamut
to the printer gamut, extra performance can be squeezed out of the method by
computing the transformation for each image. The image gamut, being a subset
of the monitor gamut, will hopefully have less outliers. The saturation loss can
thus be minimised. Of course this is more computationaly expensive.

University of Manchester 77
6 Usage of colour

6.1 When to use colour

Used wisely, colour can greatly add to the usefulness, clarity and impact of a
graphic. Used badly, however, it serves only to confuse and obscure. Colour can
be used:

• to distinguish objects
• to show relationship and connections between objects
• to display additional information without an increase in dimensionality
• to make clear the 3D form of an object
• to make a graphic more interesting to the viewer
• to direct attention to parts of a graphic

If colour is being assigned some coded meaning, for example in status displays or
user interface design, the number of colours should be strictly limited. Many
studies have shown that no more than six or seven colours should be used in this
fashion, and they should be clearly distinguishable. Furthermore, it is preferable
to add redundant information such as size or shape to reinforce the distinction,
as shown in Plate 42

If a continuous colour scale is being used to display the value of some variable –
such as temperature or stress – over a surface then using more colours gives a
finer gradation and enables small details to be seen. In this case, 256 colours
from an 8 bit display may well be insufficient.

6.2 Selecting a colour model

For device independence, use a CIE colour model such as L*C*Huv, or a model
such as Munsell for which CIE tristimulus values are available. If the chosen
output device does not directly support CIE colour specification, colours can often
be converted to the native colour space of the device. Appendix B shows how this
is done for an additive RGB device.

For accurate work, where an exact colour match is important, use a calibrated
monitor and a viewing cabinet, with CIELUV or more specialised models such
as Hunt ACAM.

For particular applications - textiles, architecture, graphic arts - use the appro-
priate specialised colour model which is conventionally used in that application
area. For example, CIELAB, Coloroid, or Pantone respectively.

University of Manchester 79
 To mix device independent colour, use a colour model with a polar coordinate sys-
tem to give a hue wheel. For example, L*C*Huv or TekHVC.

To mix distinguishable but device dependent colour, use a polar model such as
HSV or HLS.

Use RGB if it is all that is available, but consider selecting from HSV and con-
verting to RGB.

Do not use CMY directly. The number of variables which must be altered is too
large.

6.3 Colour schemes

A range of related colours used together give a unified, uncluttered look. A poor
selection of colours can look confusing or garish and may contribute to eyestrain
if viewed for long periods of time. To some extent, the selection of related colours
is a subjective process; however help can be obtained from both artistic conven-
tions and colour science. In many cases, the empirical guidelines from the artistic
world can be explained in terms of colour science.

6.3.1 Using artistic conventions

Complementary colours are those on opposite sides of a colour wheel. Using
complementary colours produces a busy, attention-getting display. To avoid look-
ing garish, the chroma should be reduced for large areas. It is often useful to use
one colour or group of similar colours for large areas, so that their complemen-
tary colours will stand out when used in small amounts as highlights or accents.

Different colour models use different spacings of colours round a hue wheel, so
the exact colour which is found to be the complement of another varies with the
colour model. A true perceptually uniform scale would give the correct colour.

Determining complementary colours can be readily done by eye. Simply stare fix-
edly at a small patch of colour on a black background for a minute or so. Looking
at a well illuminated white surface will produce an after image in the comple-
mentary colour. Plate 43 can be used to try this out. Note that the precise colour
obtained from an after image depends on the illuminant used.

After images are caused by the photosensitive pigment in the cones becoming
bleached as a continuous high chroma colour stimulus is applied. Staring fixedly,
without moving the eye, keeps the image on the same cells in the retina. In the
resting cell, used pigment is replenished. By not allowing this to take place, the
pigment in each cone type is depleted in proportion to the degree to which that
colour excites each cone type. For example, a green stimulus will bleach M cones
the most. Looking at a white surface will give the illusion of a pink colour until
the pigment is replenished.

A graphic which incorporates many intense unrelated colours will look cluttered
and confusing; there is no single point of focus. Using groups of related colours

80 Computer Graphics and Visualisation

and using high chroma colours sparingly for accentuation avoids this effect and
gives a focused, controlled and professional look.

If particular colours are to have individual meanings, these should be clearly ex-
plained and the colours readily distinguishable. Some colours have conventional
meanings which are widely – if not universally – understood. For example, red is
associated with action, excitement, danger, heat and stop. Such meanings are
overloaded and may be contradictory. They may also be specific to a particular
culture.

When a smooth range of colours is to be used, it is useful to incorporate existing

meanings, especially those used unambiguously by a clearly defined group. For
example, in medical imaging the convention is that red denotes normal tissue
and blue, diseased tissue. In cartography, a range of dark blues shading to light
blues and white represents progressively shallower seas; yellows, greens and
browns represent increasing land height, culminating in purple and white for
mountain tops.

6.3.2 Using colour science

Don’t have blue and red together. Don’t use blue as a foreground colour, where
shape must be distinguished; however it makes a good background. Why: chro-
matic aberration in the eye makes it impossible to fully focus on red and blue
simultaneously as seen in Plate 44. The eye will tire from continual re-focusing,
and settle on a lens position where neither colour is fully in focus. When using
blue as a background, it has no fine shape so gentle blurring is unobtrusive. The
phenomenon of stereopsis gives the appearance of depth.

Don’t have fine detail in blue or red on dark coloured backgrounds. Why: the
photopic luminous efficiency curve is sharply peaked in the yellow and green
part of the spectrum. Colours at the spectral extremes will appear much darker
than yellows and greens at the same measured light power level. Similarly, yel-
lows and greens on a light background will have low contrast and thus be
difficult to see.

Don’t use blue or violet for small moving shapes such as mouse cursors. Why: S
cones have a slower response than M or L cones. Therefore they cannot detect
rapid changes in position of blue and violet objects. The density of S cones in the
fovea is much less than the density of M and L cones. Therefore, the spatial reso-
lution for blue objects is much less than for other colours ( a fact which, as we
have seen, is made use of by subsampling in video encoding)

Don’t rely on red/green discrimination to convey important information. Why: a

significant proportion of your audience will have reduced or missing sensitivity
to red/green differences.

Do use perceptually uniform colour spaces to construct colour scales. Why: Colour
scales with perceptual jumps can give a false impression of spurious detail. Areas
of little colour change can mask details. A perceptually linear colour scale facili-
tates estimates of the displayed parameter.

University of Manchester 81
6.4 Interpolation
The colour space chosen affects how colours are interpolated. This has bearing on
the production of colour scales for visualisation. Perceptually uniform spaces are
to be preferred to avoid discontinuities or distortions of scales. Polar coordinate
spaces are often easier to work with than Cartesian spaces.

In addition to straight linear interpolation, it may be useful to construct a colour

scale along a curve through some colour space. Some visualisation systems allow
curves to be used in this manner. Examples of colour scale creation using this
method are shown using AVS (Plate 45) and Explorer (Plates 46 and 47 ). Both
examples use the HSV space, called ‘HSB’ in AVS. Plate 45 also clearly shows
the lack of perceptual linearity of hue in HSV. Although the curve through HSV
space is smooth, there are sudden perceptual jumps; for example between yellow
and orange.

In Plate 47, the red curve is hue, moving from red (0°) at the bottom to violet
(360°) at the top. The green curve is saturation, and the cyan curve is value. The
small filled squares represent points that can be moved, through which the curve
passes. (These are termed knots, and will be met in the Curves and Surfaces
module). The open square on the green curve represents the tangent to the curve
at the point being edited.

The oranges are desaturated to give browns, and the purples made lighter. This
colour scale was chosen to illustrate the method rather than be an example of
good practice.

Non linear colour interpolation may be used as a form of depth cueing. For exam-
ple, in the representation of outdoor scenes, distant objects can be made more
blue and less saturated. This mimics the effect of atmospheric haze.

Pseudo colour can be used to enhance detail or visualise small changes. Exam-
ples of this type of application are medical imaging, geographical information
systems, and finite element post-processing.

82 Computer Graphics and Visualisation

A Gamma correction

A.1 Determining gamma

There are three ways of obtaining a value for gamma correction of a monitor:

1. Direct measurement of standard greys using a light meter or

spectroradiometer

2. Asking the monitor or tube manufacturer

3. Visual calibration

The first method can have the additional refinement of measuring the value for
each gun separately.

Some monitors have internal gamma correction in hardware, and need no fur-
ther adjustment. To detect such a monitor, refer to the manual or carry out a
quick visual calibration. The majority of monitors will, however, require gamma
correction.

A.2 Direct measurement

For simplicity, we will assume that all three guns are to be calibrated together.
The method simply consists of generating a series of test patches of known RGB
value, measuring the actual light emitted, and plotting the test value against the
measured value.

Using a large number of samples, and performing duplicate tests, helps reduce
random errors and give a more precise result. Each patch should be measured at
the same part of the screen, conventionally the centre, to minimise the effect of
misconvergence. If desired, a patch near one corner can be measured in a sepa-
rate series, and the results averaged.

Provided the phosphors are not being driven into saturation, the measured
gamma value should be much the same regardless of the setting of the bright-
ness control. Although the eye can adapt to the ambient light level, a meter
cannot, so the screen and meter should be well shrouded with heavy cloth such
as a curtain, to eliminate stray light.

Remembering that the gamma function is a power law, the input RGB value and
output light level should be plotted on a log/log scale. The spacing of the test
samples should take this into account, so that samples are evenly spaced on the
log axis.

University of Manchester 83
 If the data points cluster around a straight line, the slope of that line is the
gamma value. Significant deviations from a straight line can only be dealt with
by a lookup table.

A.3 Visual calibration

This simple method has the advantage of requiring no equipment. It relies on
visual comparison of two grey patches. Visual comparison can be quite accurate
and precise, and is after all the basis of the CIE standard observer.

The method relies on the fact that, regardless of the gamma value, white and
black are fixed points. This is shown in Figure 60; altering the gamma value only
affects the amount of curvature, not the position of the end points.

1.0
light output

0.0
0.0 1.0
input RGB

Figure 60: Three gamma curves

If a chequerboard pattern of black and white squares is displayed, the result

looks grey if the grid is fine enough. The light level will be 50% of the maximum,
white light level, because of additivity (assuming that the black is effectively
zero). This corresponds to the grey that would be obtained with R,G and B all
0·5, if the monitor had a gamma of one.

A grey colour is mixed, for example using a paint package, keeping R,G and B
equal. If the HLS colour model is available, mixing a colour with a saturation of
zero will accomplish this. Call the value of RGB which matches the chequerboard
pattern V. The gamma value is given by:

gamma = log(0·5) / log(V)

For example, if V is 0·73, the gamma is 2·2. This is shown in Figure 61.

84 Computer Graphics and Visualisation

 1.0

light output
0.0
0.0 0.5 V 1.0

input RGB

Figure 61: Matching grey with checks.

University of Manchester 85
B Monitor calibration

For precise work, each monitor should be calculated by directly measuring the
tristimulus values with a spectroradiometer. To compensate for drift in electronic
components, and phosphor ageing, this calibration is performed monthly, weekly
or even (in highly critical applications such as colour pre-press soft proofing or
dyestuff quality control) daily. Monitors are available which can auto-calibrate
themselves.

For general work, however, it is sufficient to obtain monitor tristimulus values

which are typical for that model of monitor from the manufacturer. The data re-
quired are the chromaticity values of the red, green and blue phosphors, and the
chromaticity of the monitors natural white point where red, green and blue are
all at maximum intensity.

Chromaticity values are generally supplied as x,y pairs. The first step is to calcu-
late the z components:

z=1–x–y

The phosphor chromaticities are inserted into a matrix A, where:

2 3
xr xg xb
A=4 yr yg yb 5
zr zg zb

This matrix is inverted, to give B:

B = A−1

The monitor white point is converted from chromaticity values to relative

tristimulus (XYZ) values. The absolute luminance is generally not required, Y
can be set to unity:

Xw = xw /yw
Yw = 1.0
Zw = zw /yw

These values are used to compute the column vector C:

2 3
Xw
C = B4 Yw 5
Zw

University of Manchester 87
 The elements of C are inserted into the major diagonal of a new matrix D, the
other elements being zero:
2 3
C1 0 ⋅ 0 0 ⋅ 0
D=4 0 ⋅ 0 C2 0 ⋅ 0 5
0 ⋅ 0 0 ⋅ 0 C3

The matrix to convert RGB to XYZ is then given by:

E=AD

and this is inverted for conversion in the reverse direction. Any point specified in
RGB may then be converted to CIE 1931 XYZ by:

2 3 2 3
X R
4 Y 5 = E4 G 5
Z B

88 Computer Graphics and Visualisation

C Glossary

This glossary is not presented as a set of formal or precise definitions. It is delib-

erately kept informal and colloquial. It is intended to refresh the memory when a
particular word is encountered.

adaptation Response to an overall change in light intensity. A ‘win-

dow’ of simultaneously perceivable light and dark is slid
up or down the entire range of brightness. A sheet of
white paper is seen as white regardless of the light level.

achromatic Having no chroma or saturation; a pure untinted grey,

white or black.

bandwidth The range of frequencies used by a video signal, from zero

on up. For example, a PAL signal has a bandwidth of 5·5
MHz.

brightness The amount of light emitted or reflected from something,

in absolute terms. A patch of grey on a white sheet of pa-
per would be brighter in strong sunlight than in the
shade. See lightness.

cathode ray tube Device for producing a colour or monochrome display by

inducing phophors to emit light.

colour difference Another name for the two chrominance signals used in
video encoding.

chroma (1) A measure of colourfullness, similar to saturation but

expressed relative to an area of similarly illuminated
white.
(2) In Munsell colour space, the departure of a colour
from grey; colour intensity.

chromatic aberration Inability to focus on both ends of the visible spectrum at

once, caused by the refractive index of the aqueous hu-
mour being dependent on wavelength. Leads to the
phenomenon of chromostereopsis.

chromatic adaptation Response to an overall change in colour of light. A sheet

of white paper tends to be seen as white regardless of the
colour of light, up to a point.

chromatic induction Neutral surrounds to colours reduce the perceived

colourfullness. Dark surrounds increase perceived
colourfullness. High chroma colours induce a tint of the

University of Manchester 89
 same colour into dark surrounds and the complementary
colour into light surrounds.

chromaticity Means of describing the hue and colourfullness of a sam-

ple independent of its lightness.

chrominance Originally related to the difference between a video com-

ponent signal and the system white; now simply a name
attached to the two video components C1 and C2 that
carry most of the colour information.

chromostereopsis Phenomenon of blues seeming to recede, and reds seem-

ing to come forward. Caused by feedback of muscular
changes to re-focus the eye being interpreted as distance
focusing.

colour constancy An evolutionary survival tactic. The eye and brain try to
hold the perceived colour of objects constant regardless of
slow changes in the level or colour of daylight. Only
partly successful, and less so for non-daylight illumina-
tion.

colour management system A piece of software that helps achieve screen to print col-
our matching, often by displaying the gamuts of the two
devices.

colour temperature The temperature of a theoretical black body radiator

which would produce light of the same or similar
chromaticity.

complementary colour The colour exactly opposite in hue to a given colour.

density Log of one over the reflectance. White and light colours
have a low density, black and dark colours have a high
density.

dithering Simulating more colours by printing two or more colours

close together

dominant wavelength (Of a colour). A single spectral wavelength which, when

mixed with white, would give an equivalent colour sensa-
tion.

dot gain Non-linear increase in the size of halftone dots. Has a va-
riety of causes, and is measured and corrected for in high
quality commercial colour printing.

dye sublimation Printing technique which sublimates (converts from solid

to vapour) inks to allow varying quantities of each ink to
be deposited, effectively forming a mixed colour for each
pixel. This gives a continuous tone image, with no need
for halftoning.

90 Computer Graphics and Visualisation

electromagnetic energy A form of energy which can be transmitted through a vac-
uum. Travels at the speed of light. Can be considered as a
wave or as a particle. Characterised by wavelength.

excitation purity Percentage distance of a point on a chromaticity diagram

from the white point to the spectral locus. Very roughly
corresponds to colourfullness, but better measures are
available.

fovea Small central area of the retina, containing many cones

but few rods, and with nerves and blood vessels routed
round the sides. Has the best colour discrimination and
spatial resolution.

gamut The range of colours displayable by a device; a subset of

the range of perceivable colours.

gamut mapping Matching and adjustment of colours from one device

gamut for display on a second, different device, so that it
looks as close to the original as possible.

grey component replacement Removal of the common proportion of cyan,magenta and

yellow in a colour, which would produce a grey, and sub-
stitution of an equal amount of black ink.

grid Part of a cathode ray tube which electrostatically acceler-

ates the electron beams by means of an applied voltage.

halftoning Simulation of intensity difference by varying the size of

dots on a regular grid.

hue The degree to which a sample resembles red, or yellow, or

green, or blue; regardless of how light or colourful the
sample is.

hue angle A single numerical measure of hue, generally considering

red to be 0°.

ISO International Standards Organisation, the body which

produces world-wide standards in many fields.

laser printer Printing device using a laser beam to control electrostatic

deposition of toner(s) on paper. Toner is fused to the pa-
per with heat or pressure.

lightness The amount of light emitted or reflected from something,

relative to white. A patch of grey on a white sheet of pa-
per would have the same lightness in strong sunlight and
in the shade. See brightness. Lightness takes adapta-
tion into account.

luminance Measure of how intense a light is, in isolation. To find out

how bright it appears, the intensity of other lights in the

University of Manchester 91
 surroundings must be considered. Luminance is a way of
measuring light intensity that takes into account the
photopic luminous efficiency curve.

luminous efficiency Perceived brightness of a fixed intensity of a particular

wavelength of light. For the same light level, green seems
a lot brighter than red or blue.

metamerism (1) The fact that two spectrally different samples can give
the same colour sensation, which allows the simulation of
many colours with only three primaries.
(2) A consequence of the information loss when computing
tristimulus values. Two spectrally different samples may
give the same values (and hence look identical) under one
illuminant, while giving different values (and looking dif-
ferent) under another illuminant.

opponent colours Three pairs of colours: red/green, blue/yellow, and

black/white. A given colour can only have the characteris-
tics of one colour from each pair. For example, a reddish
yellow is possible (orange) whereas a reddish green is not.

offset lithography A printing process where each halftoned colour separa-

tion is printed from a cylindrical metal plate. Called
offset because the inked plate prints onto a rubber roller,
which then prints onto the paper to give even pressure.

perceptually uniform Property of a colour space that equal distances represent

equal observed colour differences.

perfect diffuse reflector A white target that does not modify light at all, but re-
flects 100% of it at all wavelengths.

photopic vision Light adapted vision at low to high light levels, where
rods are saturated and cones provide colour information.

pixel Smallest individually addressable part of a display de-

vice.

pre-press Stage in commercial colour printing where images are

scanned, separations are done, colour balance checked,
and colour proofs made before committing to a print run.

process colour Method of printing using three subtractive colours – cyan,

magenta and yellow – to produce all colours; generally
adds black for contrast.

Purkinje shift Shift in the wavelength of maximum sensitivity to light

as vision moves from scotopic to photopic.

quantisation Mapping a large range of colours to a smaller range of col-

ours.

92 Computer Graphics and Visualisation

reflectance The amount of incident light which is reflected by some
object, as a percentage.

retina The photosensitive receptor layer in the eye that converts

light into nerve impulses.

saturation A measure of colourfullness. A dull grey has low satura-

tion, a vivid red has high saturation. Saturation is the
colourfullness of a sample relative to how bright it is.

scotopic vision Dark adapted vision, in very low light levels, using only
rods for monochromatic vision.

separation Creation of four halftoned images from an original con-

tinuous tone image, ready for process printing. Requires
the specification of screen frequencies and angles for each
process colour, and compensation for the characteristics
of the printing press. Additional spot separations may
also be made.

spectral locus The position of maximally saturated spectrally pure col-

ours in one of the CIE colour spaces, representing the
boundaries of visible light. Effectively, the gamut of the
human eye. The ends of the spectral locus are joined by
the purple line.

spectrophotometer Device to measure the CIE tristimulus values of a reflec-

tive sample; contains a light source.

spectroradiometer Device to measure the CIE tristimulus values of a light

source.

spot colour Single colour of ink, used without mixing. Many hundreds
of spot colours are available, covering a huge gamut. Gen-
erally only a few spot colours are used in a single print.

standard observer Table of experimental data predicting the amount of the

CIE primaries required to match light of any wavelength.
There are two standard observers, but the 1931 2° ob-
server is the most commonly used in practice.

subsampling A means of reducing the bandwidth of a video signal by

filtering out rapid changes in colour or intensity. Reduces
the effective resolution.

tristimulus value The amount of each of the three CIE primaries which,
when mixed, would give the same colour sensation as a
sample of a given colour.

UCS CIE 1976 uniform chromaticity scale diagram, an attempt

at a perceptually uniform presentation of colourfullness
and hue.

University of Manchester 93
under colour removal Addition of black to dark colours, to boost contrast.

video RAM Memory used by a workstation to contain the description

of the displayed image.

video luminance Originally a measure of the lightness component of a

video signal, now simply a name for the component which
carries most of the lightness information in an encoded
video signal. Not to be confused with luminance, although
for historical reasons they are both represented by the
symbol Y.

viewing cabinet A simple piece of apparatus used for colour matching of

reflective samples (fabric, photographs, etc) consisting of
a large box painted neutral grey with a standard light
source, D50 or D65, inside it.

wavelength The distance between two consecutive peaks in a regular

wave. Measured in whatever divisions of a metre are ap-
propriate. Visible light has wavelengths of between 380
and 730 nm.

waxjet Printing device which sprays coloured, molten wax onto

paper through a large number of very fine nozzles.

94 Computer Graphics and Visualisation