ÐÓ°ÉÔ­´´

ROSES are red; violets are blue . . . or are they? A red rose in yellow-green
light appears purply-pink, while a violet in green-blue light looks turquoise.
We are often unaware of such changes in the appearance of familiar objects
because the eye adapts to the prevailing light. As a result, we see colours
as we expect them to be, not as they really are. When lighting is poor,
for instance, the brain simply concentrates on enabling us to identify what
the available light allows us to see.

There are drawbacks, however. When a photographer snaps a scene illuminated
by an incandescent light, the photograph will have a distinctly yellow tinge,
although when the photograph was taken the colours looked true. This is
because the eye – but not the photographic film – adapted to the incandescent
light. The same problem occurs when a green dress is chosen from a catalogue;
a yellow paint is picked out for the kitchen; or a jacket and trousers from
different suppliers are carefully matched in a shop – they can all look
quite different when you get them home.

Reproducing colour accurately has always taxed graphic and textiles
designers. Computers allow designers to experiment with different colour
combinations and to preview the results immediately on screen, but they
do not ensure that designers and customers see the same thing. A designer
wants the colours on the screen to match those printed out on paper for
buyers to inspect or sent to another computer at the mill where a fabric
or wool is to be produced. This cannot be guaranteed yet. For a start, computer
manufacturers have not agreed on standards for the production of colours.
As a result, the colours in images vary depending on the make and model
of the screen and printer. Also, it is difficult to predict the combination
of inks needed to make an identical colour print. But the main problem is
our perception of colour.

It was Edwin Land, the founder of the American Polaroid Corporation,
who first realised that seeing colour in natural images depends on a balance
between longer and shorter wavelengths of light, rather than on the light
itself. Until his experiments in the late 1950s, classical colour theory
said that the colour we see was determined solely by the wavelength and
the intensity of the reflected light.

In 1959, Land showed that we see colour even when light of the relevant
wavelength is absent. He used two projectors to display on a screen similar
black-and-white images of multicoloured objects. When they were projected
separately, the images contained no colour. But then Land inserted a red
filter into one projector, so that it transmitted only longer wavelengths
of light, and a green filter into the other, so that it transmitted only
the shorter wavelengths. When he superimposed the images on screen, the
spectators saw not only red and green before them but every colour of the
visible spectrum. When he removed either the red or green filter, the only
colour on screen was green or red, respectively.

The way we see colour also depends on whether we are looking at a matt
or gloss surface, and on how that surface is lit – the colour of the light,
its brightness and the direction it is coming from. The presence of other
colours around the object we are looking at and the distribution of light-sensitive
cells in the retina of the eye also affect how colours appear.

The graphics and textiles industries try to get round some of these
problems by specifying standard conditions for viewing a fabric or picture.
They control the type of illumination, its direction and brightness and
the colour of the background against which the object is displayed or printed.
Experienced staff will look at samples of fabric or wool under the standard
conditions, to check that the colours are correct, before they authorise
production. But this is a costly and time-consuming business, and customers
still do not always get what they see.

Since 1987, a team of researchers in Britain has been trying to develop
a computer system to cope with our perception of colour. It has already
produced a mathematical model that enables a computer to predict how an
image on screen will look on paper or on another computer screen under a
range of viewing conditions. Now the team plans to make the model easier
to use and compatible with a wide range of computers, monitors and printers.

The team of researchers comprises scientists and engineers from the
University of Loughborough’s Technology Centre for Computer-Human Interface
Research and from Crosfield Electronics, which makes colour scanners for
the printing industry. Britain’s defunct information technology research
programme, Alvey, provided Pounds sterling 100 000 for the project in its
first three years. Backing now comes from the Information Engineering Directorate
of the Department of Trade and Industry, which is providing about Pounds
sterling 300 000 over three years. The team has been joined by Coats Viyella,
a British textiles group, based at Paisley in Scotland. Coats Viyella would
like to install a system in shops that would enable customers to look at
a screen to see how the colours in an upholstery fabric, for example, change
under different lighting conditions. The company also wants to use the model
to produce sample charts, which show buyers the latest range of colours,
on paper instead of on cloth; cloth charts are expensive to make.

According to Lindsay MacDonald, the project manager, the key to controlling
colour in computer-based systems is the method of specifying it. At present,
the description depends on whether designers are using paper or computers.
They instruct a printer what proportions of four inks – cyan, magenta, yellow
and black – to apply to the paper, or they rely on phosphors behind a display
screen to emit the required intensity of light in the red, green and blue
wavelengths of the spectrum.

But there are problems. Our perception of lightness or darkness does
not vary uniformly with the amount of light or ink present. It is also difficult
to predict the colour produced when one ink is printed on top of another:
even experienced designers use sample charts and proofs to help them. What
is needed, says MacDonald, is a better method of describing colour.

Several models exist that try to match our perception of colour. In
the 1970s, Tektronix, an American supplier of electronic displays, introduced
a system that uses what are considered to be the three most important perceptual
attributes of colour – hue, colourfulness and lightness.

Hue is what we generally think of as ‘colour’. It refers to the greenness
or blueness of a pigment, and varies with any change in the dominant wavelength
of the light reflected from an object. Colourfulness, or saturation, describes
the degree to which the hue is present – the extent to which an object is
coloured as opposed to achromatic. It is diluted as the intensity of white
light increases. Lightness refers to the brightness of a colour relative
to the white that is present in an image; it increases as the energy of
the reflected light increases. But the Tektronix model does not take account
of how colours look under different viewing conditions.

ICI recently adopted the Natural Colour System, developed in Sweden
in the late 1970s, for its Dulux range of paints. The company decided that
the system specified colour more precisely than any other one available
at the time. Architects and interior designers also favour NCS. Another
system, originally developed in the 1930s by the Commission Internationale
d’Eclairage (CIE), an organisation that specifies methods of measuring colour,
is used for making inks and dyes. The system evolved from experiments with
a group of 17 people who were asked to match sets of colour. It specifies
colour in terms of the light source, the object and the observer. The CIE
system was updated in the 1970s, and there are now two versions: one for
display monitors and another for dyes and paints.

For the basis of its work, the project team drawn from Loughborough
University and Crosfield chose a mathematical model developed by Robert
Hunt, a visiting professor of physiological optics at London’s City University.
Hunt’s model predicts what signals reach the brain when the eye sees a particular
colour under specific lighting conditions. The researchers designed a set
of experiments to test the model and improve its predictions.

The main experiment involved 10 subjects with ‘normal’ colour vision,
defined as that of about 92 per cent of the population. Researchers asked
them to estimate the hue, colourfulness and lightness of a range of coloured
patches; these were surrounded by other colours in an attempt to simulate
real scenes. Very rarely do we look at a blank wall of one colour; most
of the images that fill our retinas are complex and multicoloured. Different
colours in a scene influence each other’s appearance by enhancing the contrast
between them. For example, green leaves deepen the red of a rose – they
increase the red’s colourfulness.

The researchers changed the surroundings of the test patches to see
what sort of background interferes least with the appearance of colour.
This knowledge could improve the design of complex display panels in aircraft
or in the control rooms of nuclear power stations. They varied the type
of lighting, its intensity and the brightness of the surroundings, displaying
the patch both on a computer screen and on paper. The research, which produced
more than 43 000 estimations, provided the most information in the past
40 years on the way we perceive colour, says MacDonald. The team used the
data to test existing models of colour appearance and to refine its own,
the Hunt-Alvey Colour Appearance model, which MacDonald claims is now the
most accurate one. To improve the model, so that it makes more accurate
predictions of colour appearance, the team might choose colour experts,
such as photographers and graphics designers, as future subjects.

The researchers have also been trying to develop a method for transferring
colour between computer monitors and printers, so that they produce identical
images from the same information. First, the team had to establish the range
of colours that each device could generate. To do this they used a standard
procedure of colour research. For a printer, Crosfield created a series
of nine charts each made up of 9 by 9 samples of colour – 729 samples in
all. The proportion of yellow and magenta inks in the samples varied across
each chart from 0 per cent in the top left-hand corner to 100 per cent in
the bottom right-hand corner, and the amount of cyan varied from 0 to 100
per cent over the nine charts. MacDonald’s team ignored black; this does
not matter because black is generally added at the last stage of the printing
process (cyan, magenta and yellow by themselves produce only a brownish
black).

The researchers used a spectrophotometer to measure the amount of light
that each sample reflected at intervals of 20 nanometres throughout the
visible spectrum, from 400 to 700 nanometres. The measurements showed how
each colour is built up from different wavelengths of light of varying intensities.
The researchers then had to convert this information into a form that a
printer could interpret as colour.

ÐÓ°ÉÔ­´´s already know how each of the three types of light-sensitive
cells in the retina respond to light of a particular wavelength and energy.
Using this information, MacDonald’s team obtained the amplitude of the response
in the red, green and blue receptors of the eye for every one of its measurements.
For each of the 729 samples of colour, the team calculated a set of values,
each comprising three numbers. These matrices enabled the researchers to
describe a colour numerically so that the printer could reproduce it.

The team did the same for a monitor, using 729 samples of colour displayed
on a screen with varying proportions of red, green and blue. Instead of
a spectrophotometer, the team recorded the composition of the colours with
a telespectroradiometer, which measured the emission spectra of every sample
directly from the screen. Again, the team derived matrices that described
the colours numerically, so that the monitor could be instructed precisely
which colour to reproduce.

The researchers then developed computer software that could use the
stored information to interpolate intermediate colours made up of cyan,
magenta and yellow, or red, green and blue. The big advance, says MacDonald,
was to combine this method of describing colour on a computer monitor or
printer with the improved Hunt model, which predicts how lighting conditions
influence a colour’s appearance. In theory, this means that the team can
display an identical image from different printers and monitors and predict
how a colour’s appearance will change under different lighting conditions.
In practice, providing every make of printer and monitor with its own colour
matrices is expensive and time-consuming. One of the main aims of the new
project, which began in February, is to devise a mathematical model that
needs fewer measurements of the composition of colours, on paper or screen,
to represent them numerically for a new printer or monitor.

The work could revolutionise the designer’s task and help to ensure
that what clients and consumers see is what they get.

* * *

The science of exhibiting and renovating Old Masters

TWO questions facing art galleries around the world are how to display
paintings to their best advantage and whether a treasured Old Master should
be cleaned up. Three of Europe’s major galleries – in Britain, France and
West Germany – are seeking answers with the help of the colour appearance
model being developed by the group of researchers from Crosfield and Loughborough
University.

The three galleries want to develop a computer system that can store
and display accurate images of paintings, including their colour and surface
texture. These images will form the core of a scientific investigation into
the deterioration of paintings over long periods of time. The European information
technology programme, Esprit, has provided Pounds sterling 1.4 million for
the 30-month feasibility study, which began last July.

The Vasari (Visual Arts: System for Archiving and Retrieval of Images)
project is the brainchild of a chemist and an art historian, David Saunders
and Anthony Hamber. Saunders works at the National Gallery in London, where
he investigates colour fading and the effect of the environment on paintings;
Hamber works at Birkbeck College, London.

Using image-processing software that the pair developed with computer
scientists at Birkbeck College, they can manipulate images of paintings
on the computer screen. For instance, they can magnify portions of a painting
in great detail instead of relying on photographs that are expensive and
time-consuming to produce.

The Crosfield/Loughborough University model may help them to assess
how restoration work, such as cleaning, would affect the appearance of colours
in a painting. To take advantage of the model, restorers would clean only
tiny areas of different colours in the painting and use the spectrophotometer
to store these colours in the computer. Before any more work was done, they
could preview the ‘cleaned’ image on screen to decide if complete restoration
was desirable.

Galleries could also use the model to help them to lay out exhibitions.
The model could predict how a painting would look under different types
of illumination and with different wall colours.

Angeli Mehta is a science and technology journalist. She works for the
BBC.

Further reading Colour: Why The World Isn’t Grey by Hazel Rossotti (Penguin).
Eye, Brain and Vision by David Hubel (Scientific American Publications).