home
***
CD-ROM
|
disk
|
FTP
|
other
***
search
/
Joint Education Initiative
/
Joint Education Initiative.iso
/
dos
/
utils
/
color.txt
< prev
next >
Wrap
Text File
|
1990-07-05
|
89KB
|
1,595 lines
COLOR AND ITS USES
by Russell A. Ambroziak, U.S. Geological Survey
Color is part of our senses and some scientists do not
consider the human sensory system a scientific instrument. Some
claim that only instrument data are useful in research. What
they mean is only instrument data are quantified measurements of
reality and the mind is only useful for relative assessments of
the world around us. A color image is a 5-dimensional graph --
nothing more -- nothing less, and that makes it a powerful tool
for scientific analysis. If these scientists never drew a graph
or looked at a color image then the subject would end and there
would be little need for this presentation. We do, however, draw
graphs and create color images which cannot be analyzed by any
instrument except the human mind. We use color all the time, but
since few of us have ever been introduced to color theory we tend
not to think of it as a scientific discipline. If we are going
to use color we should spend some time trying to understand it
because the application of color theory can:
1. shorten the time necessary to produce high quality graphics
and color images,
2. increase the information content of graphics and images by
a. increasing perceptual resolution or
b. increasing the dimensionality of the data display and
3. in some cases produce analytical results which cannot be
achieved in any other way.
WHAT IS COLOR?
Every seeing person with color vision knows what color is
but no one seems to be able to describe it. While artists
understand and use color as one of the main tools of the trade
most scientists dismiss color as an unnecessary subject area to
understand. To the artist, color is a science but, to the
scientist, color is an art. This makes the scientific study of
color rather rare. Many scientists use color but, since few
understand color science, the results are based more on luck than
on design. The scientific application of color can produce
better results in less time than the typical trial and error
methods used daily by many scientists. Now that the PC on our
desk can produce 16,777,216 colors we had better spend some time
learning how to best utilize them.
Color has many definitions and it is much more complex
than the physical properties of the incoming electromagnetic
radiation, their reflection, and absorption. Color is what we
see. Color is 1 percent physical, 9 percent biological and 90
percent psychological. This may be an exaggeration but not much
of one. Before getting into the terms and parts of color theory
1
lets take a minute to understand this point. Color is not a
property of light or objects but rather it is the perception of
animals. Different animals see color in very different ways.
Color, as we know it, is only perceived by primates and it may
have something to do with the search for fruit. Regardless of
the reason for color vision it only exists in primates and is
basically the same for all primates with "normal" color vision.
I don't think we need definitions for color (since
everybody knows what it is) but here are a few. Many of these
definitions contain terms which may not be familiar but they will
be discussed later.
Color is that aspect of visual perception by which an observer
may distinquish differences between two structure-free fields of
the same size and shape, such as may be caused by differences in
the spectral composition of the radiant energy converned in the
observation. In this sense, the term color is sometimes referred
to as "perceived color" to distinquish it from color in the sense
of "psychophysical color".)Wyszecki and Stiles, 1982, p487)
Color (in the psychophysical sense) is that characteristic of a
visible radiant power by which an observer may distinguish
differences between structure-free fields of view of the same
size and shape, such as may be caused by differences in the
spectral composition of the radiant energy concerned in the
observation. Psychophysical color is specified by the
tristimulus values of the radiation power (color stimulus)
entering the eye. (Wyszecki and Stiles, 1982, p723)
Color is the attribute of visual perception that can be described
by color names : white, grey, black, yellow, orange, brown, red
green, blue, purple, and so on, or combinations of such names.
(Billmeyer and Saltzman, 1981, p 1986)
Color is any of manifold phenomena of light (as red, brown, pink,
gray, green, blue, white) or of visual sensation of perception
that enables one to differentiate objects even though the objects
may appear otherwise identical (as in size, form, or texture).
(Webster's Third New International Dictionary) Color is a
relative, conical coordinate system in Reimannian space which
makes up three of the seven dimensions of primate perception
related to vision and not dealing with location (x,y,z) nor time
(t). (Ambroziak, 1988)
If you didn't know what color was before reading these
definitions you wouldn't have gained much by reading them. If
you can see color you know what it is and if you can't see color
you will never know. Color has something to do with seeing but
not with location or time. Rather than spend a lot of effort
discussing the definition its best to try to discuss its parts
and leave the definition alone.
Just as income does not equate to savings, the spectrum
of light coming from an object does not equate to color. Savings
2
is related to income and spectra are related to color but they
are not the same. The spectrum of light coming from an object is
purely physical and quantifiable. The light enters the eye and
activates biological sensors which loosely correspond to red,
green and blue (RGB) which are also quantifiable. From this
point on, the phenomenon of color is psychological and
quantification is not possible.
The light coming from an object is the product of the
spectrum of the incoming light and the reflective properties of
the object. The cones in the eye respond with three values for
each spectrum received called "tristimulus values". These values
are the sum of the products of the incoming spectrum and the wave
length dependent observer functions for each of the three sensors
in the eye. There are an infinite number of spectra which can
produce identical tristimulus values. This produces the phenome-
non of metamerism -- colors that appear identical even though
their spectra are very different. This causes problems for color
matching using different dyes or colorants. A color may be
perfectly matched under one light source but appear very
different when the light source changes. Artists realize this
and try to paint under light from a north window as a standard.
Paintings done under artificial light may change drastically when
viewed under natural light or different artificial light and the
effect of the work may be ruined.
If this is not enough to confuse the issue of color, the
optic nerve takes the output from all of the sensors and encodes
parts of the information for transmission to the brain. At this
point much of the absolute information is lost and relative
information is enhanced. We also go from a three hue red-green--
blue (RGB) system to a four hue intensity-hue-saturation (IHS)
system. The primary sensors in the eye are red, green and blue
but the primary hues (those which do not appear to be the
combination of other hues) are red, yellow, green and blue. At
this juncture in the process, color at a point becomes more a
function of the color of surrounding points than the color
(tristimulus values) of the point itself. Colors are more
quantifiable when they are viewed against a neutral, 18 percent
grey, background then when they are mixed. I have seen trained
observers mistake a shade of orange for green on an image which
contained a considerable amount of red.
The points that need to be understood to get some handle on
the science of color are:
1. the three dimensions of color space are the conical
dimensions of intensity, hue and saturation (IHS),
2. color space is Reimannian, not euclidian which makes
distances in color space very difficult to estimate,
3. color space can usefully be described by use of a
chromaticity diagram,
3
4. absolute quantification of color by the human visual
system is not possible, and
5. color does not exist apart from its perception by an
animal's nervous system (usually human).
The "Desert Island" Experiment
To understand the IHS coordinate system, a mental experiment
called the "desert island" experiment has been described by Judd
and Wyszecki (1975). The hypothesis is that an individual with
normal color vision and no knowledge of color science is stranded
on a desert island with nothing to do. The beaches of the island
are covered with pebbles of all imaginable colors. To pass the
time, the castaway decides to arrange the many different colored
pebbles into orderly patterns.
A first step might be to separate those which are white,
grey or black from those which are red, orange, yellow, green,
blue, etc. This would be a separation of the achromatic (black,
grey or white) pebbles from the chromatic ones. The achromatic
pebbles could then be arranged from black through increasing
brightness of grey to white. Intensity, lightness, and value are
all terms used to describe this characteristic of color.
The next step might be to separate all of the chromatic
pebbles into piles of similar hue. Hues are commonly defined as
red, orange, yellow, green, blue, violet, and magenta. These
piles could then be compared to the achromatic arrangement and
further ordered by their intensities.
Finally, our observer would notice that not all of the
reds of the same intensity are the same color. Some of the
bright red pebbles would be tomato red and others would be pink.
Both pink and red are of the same hue and can be of the same
intensity, but they are obviously not the same color. This
quality is referred to as saturation, chroma or colorfulness.
If a large number of pebbles are arranged in a large
three-dimensional array so that the color differences between
them are nearly constant, our observer would have constructed the
Munsell color space (Wyszecki and Stiles, 1982).
We can liken the Reimannian color space to the Euclidean
cone. The three dimensions of color space are the intensity, hue
and saturation. The conical-like shape arises from the way the
three axes affect each other's resolution.
Intensity, bounded on only one end at black, is at the
point of the cone. The color black has only intensity, which is
equal to zero, with no hue or saturation possible. The intensity
axis extends in some direction toward infinity with a saturation
of zero at all points on the axis. This axis of intensity, which
is black and proceeds in the direction of grey, then white, is
orthogonal to both saturation and hue. Saturation is measured by
4
the distance from the intensity axis to the edge of the cone,
where saturation is equal to one. The geodesic distance covered
by a saturation change from zero to one is a function of the hue
and intensity. The greater the intensity, the greater the
geodesic distance along hue to the edge of the cone.
Hue, which is measured by the direction of the color from
the intensity axis, can be represented by an angle. Hue is
unbounded in the sense that there is no point where hue stops
naturally. Rather, it is best represented as the cross-sectional
circumference on the cone with the hues constantly fading into
one another. The result can be thought of as a Euclidean,
conical coordinate system in which there are ellipsoids of
various size and orientation that give the size of a jpcd (just
perceptible color difference) in any direction. Convenient
labels on the hue axis are the colors red, yellow, green, cyan,
blue and magenta marked at 60 degree intervals. The geodesic
distances between these colors varies, but since there is no
attempt to have distances represented by Euclidean distance on
the diagram this is not a problem. The actual distances are
alluded to by the ellipsoids.
Continuing with the desert island experiment, if a color-
blind scientist were now to arrive on the island with instruments
which measured the amounts of red, green and blue light reflected
by each pebble, he would be confused by our first observer's
choice of arrangement. The scientist might become so confused
that he would build his own scientific arrangement to show the
first observer where she went wrong.
The most logical arrangement for the color-blind
scientist to use would be to chose one of the three primary
colors, such as red, and form piles of pebbles which all
reflected the same amount of red. These piles could be arranged
in a sequence of increasing red reflectance. Each of these piles
could be further arranged into a two-dimensional array of
increasing blue in one direction and increasing green in a
direction at right angles to the blue axis. The resultant arrays
could then be stacked, and an alternate three dimensional color
arrangement could be made.
When the scientist's work was finished and our first
observer is called to see the "correct" arrangement of colored
pebbles, she would be just as confused as the color-blind
scientist was when he saw her arrangement. There would be a
pattern of some sort in the arrangement of the scientist's
pebbles, but the pattern would not be recognized for what it is.
If the scientist had red, green and blue filters through which
our first observer could view the RGB color coordinate system,
she would immediately see why he arranged them as he did.
Through the three filters, the monochromatic intensities would
appear as planes of equal intensity which are perpendicular to
each of the three axes. As obvious as the arrangement appeared
through the filters, the arrangement would still remain confusing
without the filters.
5
People can see patterns of IHS naturally but require
instruments or filters to see RGB patterns. If images are to be
viewed by people, then it is the IHS patterns which will be seen
and not the RGB patterns. More about this later. It is a key
point.
Color Space is Reimannian
We measure distance in color space by just perceptible
color differences (jpcd). These are computed from the results of
color matching experiments by trained observers with "normal"
color vision. A test color is displayed and the observer tries
to match it by combining primary colors. The standard deviation
of the errors is a consistent measure of distance in color space.
It defines the distance one must move to be able to detect a
difference one out of three times.
When color is represented in Euclidean space, the color
distance between two neighboring points in color space is given
by
(ds * ds) = (dU1 * dU1) + (dU2 * dU2) + (dU3 * dU3)
where: s = distance between points which are just perceptibly
different
U = tristimulus values of the eye.
An alternate form can be assumed which has different properties
than ordinary Euclidian space (Wyszecki and Stiles, 1982)
(ds * ds) = g11*(dU1*dU1) + (2*gl2*dU1*dU2) + (2*dU2*dU2) +
+ (2*g23*dU2*dU3) + g33*(dU3*dU3) + (2*g31*dU3*dU1)
where: g = metric coefficients which are continuous
functions of U so that ds > 0 and are derived
from the variance/covariance matrix of color
matching errors.
The space in which ds is the distance element is known as
three-dimensional Reimannian space. If it is possible to
transform the tristimulus coordinates (U) into coordinates V so
(ds * ds) = (dVl * dV1) + (dV2 * dV2) + (dV3 * dV3),
that then Reimannan space is the same as Euclidian space. This
reduction, however, is not usually possible,and in the case of
color transformation, it certainly is not.
The importance of this concept lies in its ability to
describe geodesic lines between two points in color space. In
color space, the path between two points which contains the
fewest just perceptible color differences (jpcd) is the geodesic
line. In Euclidean space, these are straight lines but, in the
case of color space, they are generally curved which means color
perception space is not Euclidean (Wyszecki and Stiles, 1982).
6
Although the perceptual color coordinate system (IHS) can
be mapped into Euclidean color space (RGB), it cannot be done in
such a way that distances between points are independent of
location. In order to map IHS into Euclidean space so the
relative distances are independent of location, six-dimensional
Euclidean space is required. If fewer dimensions are used, gaps
will appear in the space. Unfortunately, the concept of six-
dimensional space is of little value to assist in visualization
of the relative position of colors.
A glance at table 1 will illustrate this point. A
triangle can be defined by three colors such as yellow, green,
and white. Now try to draw this triangle using the distances on
table 1. The distance from yellow to green is 73.4 jpcd (just
perceptible color distance) while the sum of the other two sides
(yellow to white and white to green) is
26.7 + 40.7 = 67.4
which is shorter than yellow to green.
---------------------------------------------------------------
KEY COLOR DISTANCES FOR CRT IMAGES
---------------------------------------------------------------
IHS Variable CRT Colors*
--------------------------- ----------------------------------
R-Y Y-G G-C C-B B-M M-R
---- ---- ---- ---- ---- ----
Changing Hue 83.0 73.4 31.0 73.1 130.6 59.2
(saturation maximum)
R-W Y-W G-W C-W B-W M-W
---- ---- ---- ---- ---- ----
Changing Saturation 78.2 26.7 40.7 49.7 103.3 67.1
(hue constant)
---------------------------------------------------------------
color x y color x y
--------- ----- ----- --------- ----- ------
* R = red = 0.62 0.345 Y = yellow = 0.44 0.463
G = green = 0.26 0.58 C = cyan = 0.205 0.165
B = blue = 0.15 0.08 M = magenta = 0.385 0.213
W = white = 0.333 0.333
---------------------------------------------------------------
Table 1
---------------------------------------------------------------
The Chromaticity Diagram
It is possible, however, to map two IHS dimensions into
three RGB dimensions without distortion. This means that, if one
of the IHS coordinates is held constant, the resulting two-dimen-
sional sub-space can be mapped in three-dimensional RGB space.
These diagrams are as difficult to construct as they are to use
7
and are, therefore, of limited value. Measurement of color
differences must remain mathematical once the values of "g" are
determined. Much of the description of color is done in
Euclidean space, but one must always be aware that it is not an
accurate description of color distances, just as a flat map of
the world is not an accurate description of the distances between
features on the surface of a planet.
One representation of IHS space in RGB is the
chromaticity diagram which holds intensity constant over the
surface if intensity is considered to be the sum of the
tristimulus values. The chromaticity diagram is flat and,
therefore, cannot be a true representation of the color space.
The major problem with this diagram is that it is in RGB space
with no attempt made to make relative distances between colors in
IHS space consistent. With all of its flaws, it is still a
useful tool to understand the mapping of RGB into IHS, and it
provides insight into the use of color in remote sensing.
The chromaticity diagram uses chromaticity values to
describe hue and saturation for a constant intensity. These
chromaticity values are the tristimulus values divided by the sum
of all three tristimulus values.
x = Ul / (Ul + U2 + U3 )
y = U2 / (Ul + U2 + U3 )
z = U3 / (Ul + U2 + U3 )
Since the sum of the three chromaticity coordinates is unity,
only two of them need be defined to specify the color. The
intensity of the color is not specified by the chromaticity
coordinates, and the magnitude of U2 is usually also specified in
order to identify the actual colors on the diagram.
The description of two-dimensional Reimannian color space
for the chromaticity diagram has been attempted by many research-
ers. The most commonly used is that of MacAdam (1942) as
analyzed by Silberstein and MacAdam (1945). Observers were shown
colors at various points on the chromaticity diagram and asked to
match the color. The standard deviations of the errors of these
color matches can be used to determine the coefficients of
Reimannian space.
g11 = 1 / [ (sx * sx) * (1 - r * r) ]
gl2 = r / [ (sx * sx) * (sy * sy) * (1 - r * r) ]
g22 = 1 / [ (sy * sy) * (1 - r * r) ]
where: g = metric coefficients
s = standard deviation
r = correlation coefficient of x and y errors.
8
As with three-dimensional Reimannian space, distances are
measured using metric coefficients.
ds * ds = g11*(dx * dx) + g12*(dx * dy) + g22*(dy * dy)
The MacAdam metric coefficients can be represented on the
chromaticity diagram as ellipses of equal uncertainty of recog-
nizing color differences. The distance across the ellipse in any
direction is proportional to the distance one must move on the
chromaticity diagram to obtain a just perceptible color differe-
nce (jpcd). These calculations are performed for 25 different
points on the chromaticity diagram and expressed as geometric
constants of the ellipses (table 2). The major semiaxis is given
by "a", the minor semiaxis by "b", and the angle between the x-
axis and the major semiaxis by theta where theta is less than 180
degrees.
----------------------------------------------------------------
MACADAM ELLIPSES OBSERVED AND REGRESSION APPROXIMATION
----------------------------------------------------------------
Color Cente Observed Regression
------------------- --------------------- -------------------
N x y 1000a 1000b theta 1000a 1000b theta
--- --------------- ------- ------- ----- ------ ------ -----
1 0.160 0.057 0.85 0.35 62.5 0.82 0.35 62.7
2 0.187 0.118 2.2 0.55 77.0 1.94 0.50 76.0
3 0.253 0.125 2.5 0.50 55.5 2.93 0.56 55.7
4 0.150 0.680 9.6 2.3 105.0 9.58 2.30 104.8
5 0.131 0.521 4.7 2.0 112.5 4.70 2.03 114.1
6 0.212 0.550 5.8 2.3 100.0 5.97 2.28 99.6
7 0.258 0.450 5.0 2.0 92.0 4.73 1.95 88.4
8 0.152 0.365 3.8 1.9 110.0 3.80 1.85 107.4
9 0.280 0.385 4.0 1.5 75.0 3.87 1.55 77.3
10 0.380 0.498 4.4 1.2 70.0 4.39 1.26 72.8
11 0.160 0.200 2.1 0.95 104.0 2.34 1.03 105.1
12 0.228 0.250 3.1 0.90 72.0 2.63 0.79 73.9
13 0.305 0.323 2.3 0.90 58.0 3.01 1.11 64.9
14 0.385 0.393 3.8 1.6 65.5 3.67 1.48 65.8
15 0.472 0.399 3.2 1.4 51.0 3.25 1.31 43.5
16 0.527 0.350 2.6 1.3 20.0 2.74 1.36 24.0
17 0.475 0.300 2.9 1.1 28.5 2.55 1.23 26.6
18 0.510 0.236 2.4 1.2 29.5 2.64 1.15 26.6
19 0.596 0.283 2.6 1.3 13.0 2.52 1.28 12.4
20 0.344 0.284 2.3 0.90 60.0 2.51 0.89 53.9
21 0.390 0.237 2.5 1.0 47.0 2.32 0.82 42.5
22 0.441 0.198 2.8 0.95 34.5 2.85 1.03 35.8
23 0.278 0.223 2.4 0.55 57.5 2.37 0.56 58.5
24 0.300 0.163 2.9 0.60 54.0 2.76 0.58 50.5
25 0.365 0.153 3.6 0.95 40.0 3.43 0.95 43.7
----------------------------------------------------------------
Table 2
----------------------------------------------------------------
9
Qualitative vs. Quantitative
One strength of the human mind is that it can process
qualitative data. The human mind cannot quantify anything though
vision which makes it useless for quantification of the
information contained in images, but for the analysis of
qualitative information it is unequaled. The assumption often is
that the mind can quantify, but it cannot, and confusion results.
The mind can only see relatives, and any attempt to make the mind
quantify will be defeated. This is the basis for many optical
illusions. Qualitative perception means that the appearance of a
single pixel is more a function of the pixels surrounding it than
the quality of the pixel itself.
The human mind, in the case of color, does not even receive
quantitative data to analyze. The eye encodes color information
in such a way as to enhance differences between adjacent colors
at the expense of the measured magnitude of either color. The
perceived color of a pixel is more a function of the color of the
pixels surrounding it than it is a function of its own
quantitative color as measured by instrument.
The cones of the eye which sense color appear to transmit
signals to the brain which are a function of the brightness of
the light striking them as well as of the light striking
neighboring cones. Although the functional relationship s not
completely understood, the form of the function can be
approximated by describing the limiting cases. The purely
quantitative case would be for each cone to transmit a fixed
frequency pulse for each retinal illuminance value. The totally
qualitative case would be for a cone always to transmit the same
frequency if surrounding cones where exposed to identical retinal
illuminance. The true response lies somewhere between these two
extremes but experiments indicate that the true function is
closer to the purely qualitative case than to the quantitative
case.
A hypothetical function for qualitative vision is postulated
here and tested by visual inspection of the results. Two
identical discs are divided in half by a line through the center
and colored white on one side and black on the other. When they
are spun rapidly they will both appear uniformly grey because the
eye, with its limited response time, will integrate over the
entire diameter of the disk. If the angle of the black is
increased from 180 degrees to say 190 degrees for the inner 1/2
of the disk, the center will appear uniformly darker than the
outer portion of the disk when spun because there is now more
black in the inner portion of the disk.
The hypothetical function is now applied to the disk with
two different shades of grey and a second disk is made using this
function, it is then spun rapidly and compared to the first disk.
10
The hypothetical function is:
X+r Y+r
I*(X,Y) = 100N / [ {I(x,y)/[((x-X)*(x-X))
x=X-r y=Y-r
+ ((y-Y)*(y-Y))]}] (1)
excluding the point x = X and y = Y
X+r Y+r
where: N = I(X,Y) {1/[((x-X)*(x-X))
x=X-r y=Y-r
+ ((y-Y)*(y-Y))]}
I*(X,Y) = ganglion cell output signal response
I(x,y) = retinal illuminance
X,Y = retinal coordinates of the cone
directly over the ganglion cell being
estimated
x,y = retinal coordinates of the cones
surrounding the cell being estimated
r -- defined below
Equation (1) modifies the ganglion cell output resulting
from the stimulus of a single cone by considering all cones in a
square of dimensions 2r x 2r centered around the cone in
question. This function was created here to illustrate a point
and is not meant to explain the true working of the eye. The
result, if normalized in such a way that all cones in the square
have the same retinal illuminance, is 100. This is totally
qualitative since the resultant magnitude of "I*" is dependent
only on the relative magnitudes of "I" surrounding the cone and
is totally independent of the absolute magnitude of "I" of the
cone being estimated. The magnitude of "I*" is reduced if
surrounding cones have a higher magnitude of "I", and the
reduction, or suppression, is proportional to the illuminance and
to the inverse of the square of the distance to the suppressing
cone.
Examinations of the retina in primates show the existence of
nerve cells which appear to connect cones horizontally (Wyszecki,
1982). Although the true function of these cells in not known,
it is certainly possible that they do act in a manner similar to
that described by equation (1).
If the eye is a quantitative instrument, then each of the
two disks would appear to be colored differently and would appear
as they are. Disk 1 would appear to contain two uniformly grey
sections, one slightly darker than the other, and disk 2 would
appear to be the same color most of the area with some shading
near the mid-point of a radius line. When the disks are
stationary this is exactly what is seen. While spinning, the
11
shading of disk 2 is invisible and both halves look uniform and
of different brightness.
---------------------------------------------------------------
HYPOTHETICAL GANGLION CELL OUTPUT RESPONSE
---------------------------------------------------------------
X I(x,y) I*(X,Y) I**(X,Y)
Retinal I = I(x,y) I = I*(X,Y)
Illuminance (r = 5) (r = 5)
-------- ------------- -------------- ---------------
Column 1 Column 2 Column 3
-------- ------------- -------------- ---------------
I 90 100.0 100.0
* * * *
20 90 100.0 100.0
21 90 100.0 100.0
22 90 100.0 100.1
23 90 100.0 100.3
24 90 100.0 100.7
25 90 99.5 100.4
26 90 98.7 100.0
27 90 97.6 99.4
28 90 95.8 97.9
29 90 91.9 92.7
30 110 107.7 107.2
31 110 103.7 101.9
32 110 102.0 100.5
33 110 101.1 100.0
34 110 100.4 99.7
35 110 100.0 99.4
36 110 100.0 99.8
37 110 100.0 99.9
38 110 100.0 100.0
39 110 100.0 100.0
* * * *
59 110 100.0 100.0
---------------------------------------------------------------
Table 3
---------------------------------------------------------------
The results of this experiment indicate that the retina of
the eye is more of a qualitative than quantitative instrument.
When coupled with the action of the iris, which adjusts to
maintain constant scene retinal illuminance, the eye must be
thought of as totally qualitative. If the signal transmitted
from the eye to the brain is qualitative and if it must be
decoded by the brain to produce what is called vision, then
vision must also be qualitative. If the mind never receives the
quantitative information, there is no way for it to analyze
quantitative information.
12
The preceding experiment used grey light to show the
qualitative nature of the eye. In grey light, the relative
intensities of each of the three primary colors are held constant
and are nearly equal. If the cones respond in a particular way
to light when all three types of cones are stimulated
identically, one might postulate that they will react in the same
manner when stimulated individually. To test this hypothesis,
the experiment was repeated changing only one primary color along
the x-axis of the two disks.
The results of the color test of the qualitative nature of
the eye are identical to the intensity test except it is hue,
rather than intensity, which appears to change between the two
halves of the disks. For the color test, each of the three
primary colors is treated differently. Blue is set to zero and
green is set to 100 everywhere. Red is varied in the two disks
in the same way grey was in the first experiment. The result is
again to similar disks when spinning.
This experiment proves that, regardless of how the eye-mind
system actually sees color, the color of a point on an image is
as much, or more, a function of the color of surrounding points
as it is a function of the color of that particular point.
Colors on an image are qualitative and any attempt to make them
appear quantitative will likely be defeated by the eye.
Is There Color If No One Sees It?
The perception of color is a complex mix of physics,
physiology and psychology consisting of (1) a light source, (2) a
reflecting surface, (3) the eye, (4) the optic nerve, and (5) the
mind. Each of these five parts influences what is called color
and the way it is perceived. For satellite image analysis, there
are at least two light sources -- the sun which illuminates the
original scene and the light which illuminates the image. It is
the image which is of interest to the image analyst, but its
connection to reality must be kept in mind at all times. In the
case of soft copy images (images on a CRT), there is no
reflection from an image. The phosphors on the color monitor
give off visible light directly.
Regardless of the source of the light, color vision begins in
the eye where three types of cone receptors respond to light of
various wavelengths in different ways. For light with a wave
length of 380 to 770 nm, two or more of these cones are sensitive
at each wave length over most of the visible range (Wyszecki and
Stiles, 1982). The sensitivity of each type of cone at each wave
length is known as the observer functions, x, y, and z for red,
green, and blue, respectively. The product of the spectral power
distribution of the incoming light and the observer functions
integrated over the visible range of the spectrum result in three
numbers, called tristimulus values, which describe the response
of the eye to a particular color. These values define the color
as seen by the eye and form the basis for color space definitions
(Billmeyer and Saltzman, 1981).
13
What the retina receives is encoded by the optic nerve for
transmission to the brain. At this point relative measures are
enhanced at the expense of absolutes and we go from three to four
primary hues. The processing n the brain is complex and poorly
understood but some things are known. The brain does not process
all of the incoming information, but rather selects information
according to its needs and experience with the type of scene
being analyzed. When you say you don't notice things that
surround you every day its because the brain does not process
them. The brain is looking for changes and unusually shapes in
the scene and color makes up 3 of 5 dimensions we work with.
Studies of the activity of cells in the brains of monkeys
while they were trying to correctly identify the shapes of red or
green symbols indicates that only the color of interest was being
processed by the brain. If the green symbol was the one of
interest the brain actually turned off the red signal to reduce
the amount of data that needed to be analyzed. This is why we
can so quickly spot objects of a given color in a scene of many
colors.
In this process of analysis all objects seem to maintain
their color. We regard objects as possessing color even though
much of the variation in tristimulus values comes from the
changing light source. This is because it is the object we must
identify to survive and the brain adjusts the incoming stimulus
to preserve the color of the objects in the scene. White paper
always look white regardless of the light source.
Each type of animal sees color in a very different way
according to the needs of the species. Color may be related to
the incoming light but it is not color until it is processed by a
brain. This is part of the reason that it is so difficult to
define color; color is what you see color to be.
SOME USEFUL DEFINITIONS
Color
that aspect of visual perception by which an observer may
distinguish differences between two structure-free fields of view
of the same size and shape, such as may be caused by differences
in the spectral composition of the radiant energy concerned in
the observation. In this sense, the term color is sometimes
referred to as "perceived color" to distinguish it from color in
the sense of "psychophysical color". (Wyszecki and Stiles, 1982,
p487)
(in the psychophysical sense) that characteristic of a visible
radiant power by which an observer may distinguish differences
between two structure-free fields of view of the same size and
shape, such as may be caused by differences in the spectral
14
composition of the radiant energy concerned in the observation.
Psychophysical color is specified by the tristimulus values of
the radiation power (color stimulus) entering the eye. (Wyszecki
and Stiles, 1982, p723)
the attribute of visual perception that can be described by color
names: white, grey black, yellow, orange, brown, red, green,
blue, purple, and so on, or combinations of such names.
(Billmeyer and Saltzman, 1981, p186)
any of manifold phenomena of light (as red, brown, pick, grey,
green, blue, white) or of visual sensation or perception that
enables one to differentiate objects even though the objects may
appear otherwise identical (as in size, form, or texture).
(Webster's Third New International Dictionary).
a relative, conical coordinate system in Reimannian space which
makes up three of the seven dimensions of primate perception
related to vision and not dealing with location (x,y,z) nor time.
(Ambroziak, 1988)
hue
the attribute of color perception denoted by blue, green, yellow,
red, purple, and so on. (Wyszecki and Stiles, 1982, p487)
the property of color most like the colors of the spectrum or
that property of color which does not involve the apparent
addition nor subtraction of white or black. (Ambroziak, 1988)
intensity
the attribute of a visual sensation according to which a given
visual stimulus appears to be more or less bright; or, according
to which the area in which the visual stimulus is presented
appears to emit more or less light. (Wyszecki and Stiles, 1982,
p487)
The attribute of color which appears to be reduced by the
addition of black. (Ambroziak, 1988)
observer functions
the sensitivity as a function of wavelength of each of the three
color sensors in the eye (Ambroziak, 1988)
metamerism
color stimuli with the same tristimulus values but different
spectral radiant power distributions. (Wyszecki and Stile, 1982,
p184)
15
colors which look the same to the eye but different to
instruments. (Ambroziak, 1988)
saturation
the attribute of visual sensation which permits a judgement to be
made of the degree to which a chromatic stimulus differs from an
achromatic stimulus regardless of their brightness. (Wyszecki
and Stiles, 1982, p487)
the attribute of color which appears to be reduced by the
addition of white. (Ambroziak, 1988)
tristimulus values
the amount of the three primary color stimuli required to give by
additive mixture a color match with the color stimulus
considered. (Wyszecki and Stiles, 1982, p723)
the relative amount of red, green and blue seen by the human eye.
(Ambroziac, 1988).
unique (or primary) hues
hues that cannot be further described by the use of the names of
the hue names other than its own (also referred to as unitary
hue). There are four unique hues, each of which shows no
perceptual similarity to any of the others; they are: red,
green, yellow, and blue. The hueness of a light (color stimulus)
can be described as a combination of two unique hues; for
example, orange is yellowish-red or reddish yellow. (Wyszecki
and Stiles, 1982, p487)
hues which do appear to be combinations of other hues.
(Ambroziak, 1988)
REFERENCE
Ambroziak, Russell A., 1986: Real-Time Crop Assessment Using
Color Theory and Satellite Data, University of Delaware, Ph.D.
Dissertation, 205 pp.
Billmeyer, Fred W. and Max Saltzman, 1981: Principles of Color
Technology, New York: John Wiley and Sons, 240 pp.
Judd, Deane B. and G. Wyszecki, 1975: Color in Business, Science
and Industry, New York: John Wiley and Sons.
Wyszecki, Gunter and W. S. Stiles, 1982: Color Science:
Concepts and Methods, Quantitative Data and Formulae, New York:
John Wiley and Sons, 950 pp.
16
HOW DO YOU USE COLOR?
A simple answer is that you use color like any other
graph. Color can be thought of as another way to graph
information and what is true for graphs is true for color images.
The main use of a graph is to see shapes and relationships not to
measure or quantify. The equations or data used to create the
graph contain the quantitative information but the graph is
created for the mind to do qualitative analysis of shapes and
forms. Color can do the same thing and give a 5-dimensional
graph.
If color is so important, why doesn't everybody learn
about it and use it? A partial answer to this question is the
lack of available control over a color system by scientists for
analytical work. To do reasonable analytical color work you need
a minimum of 256 colors taken from 16,777,216 palette. Until a
few years ago, this was only possible with equipment costing
hundreds of thousands of dollars, and the use of these systems
required a good working knowledge of color theory. For the most
part, color use came from photography, and digital imagery
applications of color were based on photographic presentations
such as false color-IR or on pseudo-color.
Color control was in the hands of the print, paint, and
fine art community and color scientists supported these
industries. Hard copy of color is still extremely difficult to
produce for the physical scientist but the new color boards for
PC's have opened up a new dimension for analytical work using
color as a tool is soft copy for a few hundred dollars.
There are two basic ways of using color which come under
a variety of headings, but they boil down to the display of
results of analysis and the analysis of data. The use of color
for the display of data includes color tagging, color coding,
color slicing and pseudo-color. The use of color for analysis
includes false color, the ACCS (Ambroziak Color Coordinate
System), some applications of IHS and RGB to raw data, and almost
all photographic applications. The most common application of
color by non-color scientists has been and continues to be the
first, that is, display rather than analysis. Almost all color
analysis by non-color scientists has been based on photographic
products.
The Use of Color for Display
On any image or graph, you have 2-dimensions to work with
(x and y), but adding color you can tag points and add a third
dimension. In order for this to work well, the colors must be
different enough from each other to be uniquely identified.
Another way to say this is that their distances in color space
must be maximized. A list of such color might be:
17
---------------------------------------------------------------
REASONABLE COLORS FOR USE IN TAGGING
---------------------------------------------------------------
# color intensity hue saturation
------ ------------- ------------- ----------- ------------
1 red max red max
2 green max green max
3 blue max blue max
4 yellow max yellow max
5 magenta max blue-red max
6 cyan max blue-green max
7 black zero none none
8 orange max red-yellow max
9 white max none zero
10 grey middle none zero
11 brown middle yellow max
12 forest green middle green max
13 pink max red middle
14 rose middle red middle
15 navy blue middle blue max
----------------------------------------------------------------
Table 4
----------------------------------------------------------------
A quick glance at these colors gives one the feeling that
they are reasonable choices for a color display. These colors
vary with the medium used to produce them, e.g., a flat bed
plotter does not have the range of color that a CRT can produce.
The problem arises when you try to add another color. It can be
done but it may be too close to one of the above colors to be
distinguishable everywhere on the graphic.
Ten of the 15 colors have maximum saturation and three
have zero or no saturation. Only two have any gradation of
saturation and both of these are red. This might be a place to
add a color, so lets try it. Yellow looks so much like white
already that mixing the two to create another color is out of the
question. Cyan is only slightly better than yellow, in the real
world, and no better on a CRT, so half saturated cyan is out,
also. Green is a little better but not enough to do much good.
Blue, magenta, and orange all have good resolution in the
direction of saturation but present other problems. Orange and
magenta tend to become confused with red when saturation is
dropped and unsaturated blue is almost undistinguishable from
cyan on most systems.
The point of all of this is to say that the order of
magnitude of the number of colors that you can use to tagging is
15 -- seven if you are using a plotter and maybe 20 if you have a
CRT, but at 20 colors, you're pushing with any display medium.
If you can display 15 values at any point, you have a color
display system which can give an x and y positional value and tag
the point with one of about 15 values. Which is a two
dimensional histogram of sorts with 15 bins each labeled by a
readily distinguishable color. This a common type of system for
18
display, since you can get all of this on a CRT with 4 bits of
data.
It is not the small number of colors which presents the
major problem for application of this system to analytical
problems but it is the lack of direction or order to the colors.
Once the colors are picked, any random order is about the same as
any other order. If you want to display pressure in psuedocolor
using the above scale you can divide the range of data into 15
chunks and color the map with ease. The problem is that higher
(or lower) values do not seem to move in any logical direction.
Is white above or below blue?
If you have developed a climate classification system and
wish to analyze the geographic extent of 15 categories of climate
psuedocolor is an ideal choice for displaying your results. You
would probably want to vary the color choices, but the basic idea
would be the same -- no direction, just tags. This application
can be used for any number of problems, such as, separating
various lines on a graph, detecting slight density changes in
monochromatic images, or mapping any type of non-sequential
categories.
The Use of Color for Analysis
The main reason for displaying data in image form for
analysis is to use the minds ability to detect non-geometric
patterns and textures. We, humans, are very good at this sort of
thing and computers tend to be less than inept at spotting and
identifying fractals. We can easily identify all kinds of
patterns on images at a glance, such as clouds, river valleys,
man-made structures, etc. Not only can we identify them, but we
can see more than the image contains. Visual perception is the
result of what you see times your experiences. If you have no
idea what your looking at, you will see very little, but it you
know a lot about the situation, you will see much more. A
weather forecaster and a non scientist looking at the GOES image
on the evening news see very different images and get different
information from it.
If you're going to do analysis you want to display as
much data as possible, and display it in the clearest manner for
the mind to interpret. You want magnitude and direction to be
preserved and you want the color axes to be orthogonal. To
reduce the amount of trial and error involved in creating an
image, it helps to understand something about the way the mind
interprets color images and then start with a suitable IHS
coordinate system.
Color has long been used for analysis by photo
interpreters using color or false color film. These products
produce images in which all of the original data are displayed
and direction is preserved. In any channel, brighter objects
appear brighter. As stated before, the human mind processes IHS
19
data while photos are RGB data so we must convert the RGB space
to IHS space in order to evaluate what the observer is observing.
RGB images usually equate color intensity with the
intensity of brightest of the three RGB values. Hue and
saturation display the ratios of the three RGB values. If two
channels are highly correlated as is the case in false color-IR
image their ratio with the third channel will appear as
saturation.
AN IHS APPLICATION CASE STUDY
The False Color IR Image
The false color infrared (IR) image, commonly used for
monitoring vegetation, is created by the computer as RGB images,
but for crop monitoring they are IHS images. Fortuitously, the
IHS coordinate system in the RGB color space of the false color
IR image is aligned with the most commonly used vegetation index.
Color IR film was created to detect foliage camouflage.
Military equipment can be painted green, but it cannot be painted
to look like foliage on color IR photos. Plants reflect much
more near IR radiation than visible, and all paints reflect
nearly equal amounts of both visible and near IR. To make the
film, all three primary colors are shifted to longer wave
lengths. Blue light is dropped, green light is recorded as blue,
red light is recorded as green, and the near IR is recorded as
red by the film. The result is an image which displays
vegetation as bright red, water as black, dark blue or cyan and
most other objects as grey or white.
The false color IR image is an RGB image which displays
each of three sensor magnitudes as a primary color. In the case
of satellite images, these sensor magnitudes are the incoming
visible and near IR radiation. For Landsat channels 4, 5, and 7
become blue, green, and red respectively. For NOAA AVHRR, there
is only one visible channel, so the blue and green are both used
to display channel I while red alone displays channel II. The
resulting images are similar because NOAA AVHRR channel I has the
same spectral range as Landsat channels 4 and 5 combined and
natural scenes contain little crop status information within the
green-red portion of the spectrum. Most of the information is
contained between the visible and the near IR.
Channels I and II of the AVHRR are ideal for monitoring
the health and status of vegetation because healthy plants absorb
visible light as a good source and reflect almost all near IR
radiation. Turgid plant cells (swelled by internal fluid
pressure) provide air-water interfaces which are highly
reflective to near IR radiation. As soon as the internal
pressure of the cells decreases, this high reflectivity falls
rapidly. The loss of near IR reflectance can be detected before
the plant shows any physical sings of wilting. After a period of
20
days, the water stress may effect the chlorophyll and visible
reflectance will increase. Reflectance varies considerably
between different types of plants and with their phenological
stages, yet, when a full cover of healthy plants is viewed
through a clear atmosphere, there can be no doubt about the
target's identity.
Obviously, the analysis of this index is a valuable tool
for the crop monitor and its relationship to false color IR
images provides insight into the initial success of research
programs designed to monitor vegetation from space and the
operational success of the current assessment programs. The
false color IR image has been used in crop monitoring without
question for at least the past decade. It is the standard method
of three channel image display; it works and it is used.
The false color IR image is actually an IHS presentation
of the Normalized Vegetation Index (NVI), and this is the reason
for its success as a crop monitoring tool. The sign of the NVI
is given by hue. All positive values are red and all negative
values are cyan on both Landsat and AVHRR false color IR images.
The value of the index is given by saturation. The diagonal,
which has an index value of zero, has a saturation of zero and
the saturation increases to one towards both axes. Scene
brightness is portrayed as intensity in a natural manner.
It is the saturation of the hues red and cyan which
contain the most valuable information in the false color IR
image. If the saturation of any pixel is compared to the
absolute value of its normalized vegetation index they will be
found closely related. The calculation of hue and saturation is
complex on a normal chromaticity diagram but when an image is
formed by three pigments or phosphors the problem is simplified
somewhat. The triangle formed on the chromaticity diagram by the
three primary colors contains all possible colors. If the edges
of this triangle, which represent the highest color saturation
possible, are defined as having an image saturation of one, the
problem becomes even simpler. In the case of red, this is good
approximation because the red phosphor has an actual chromaticity
saturation of 0.905, but the cyan on a CRT has a saturation of
only 0.414. This is not a serious problem because most of the
useful information on a false color IR image is in the red. The
cyan hues tend to be very dark when they increase in saturation
because it is only water which has index near minus one.
The image saturation and the absolute value of the
normalized vegetation index are equal at zero and one. Between
zero and one, the difference is never more that 0.10. The
normalized vegetation image is displayed as a smooth change in
saturation which is nearly the same as the vegetation index.
Although the false color IR image is an RGB image formed
from the direct assignment of three channel data to the intensity
of the blue, green, and red phosphors on a CRT, it fortuitously
is an IHS display of the normalized vegetation index. This is
21
the reason that so much information about agriculture could be
seen in the early Landsat images. It is also the reason that
image interpretation of these false color images, whether from
Landsat or NOAA AVHRR, are the prime information source in all
operational systems.
If it is the human mind which is to do the analysis of
the satellite image then the question must be asked "Is false
color IR the best type of image to analyze for satellite analysis
of large scale agriculture?" The false color IR image displays
all of the information about plant vigor in terms of saturation
changes of red (remember, cyan is for water). The path of the
color on the chromaticity diagram is a straight line from white
to red phosphor, which has a length of 78.2 jpcd (table 1). This
means that a maximum of 78 different colors can be identified,
under ideal conditions, between the best vegetation and no
vegetation at all.
Another favorite color scheme is to use green for near IR
and magenta (blue and red) for visible to make vegetation green
rather than red. The distance from green to white along the
chromaticity diagram straight line is only 26.7 jpcd or about
one-third the color resolution as false color IR. Although this
display provides more natural colors, it is rarely used for
analysis because of the poor color resolution it provides for
analyzing patterns of the normalized vegetation index.
The best image of the false color family would be one in
which the near IR (AVHRR channel II) is assigned to blue, and
yellow (red and green) assigned to the visible. The distance
from blue to white along the chromaticity straight line is 103.3
jpcd and, therefore, has 32 percent more visual color resolution
than the more common false color IR with red on the near IR
channel. The colors of this image are not pleasing and the
improvement is not appreciated by analysts who see this type of
image as strange and not worth getting used to.
SOLUTION TO THE PROBLEM
Image Color Coordinate System -- ACCS
Although the false color IR image has been enormously
useful in vegetation analysis from space, it is obvious from the
discussion of color theory and of perceptual color resolution
that it can be improved. Any improvements must keep the
favorable points of the color IR coordinate system while
improving its weaknesses.
Strengths of the false color IR image
The strong points of the color IR images are (1) its
ability to display all of the spectral information of interest
for analysis, (2) its display of the vegetation index on an axis
22
of the minds natural coordinate system, (3) the ability to
provide the information which allows the distinction between
lands, water, and surface features, and (4) the relatively low
cost of image processing. These are the positive attributes of
the color coordinate system which have made the color IR images
an integral part of all operational assessment programs and which
can afford the expense of satellite data and satellite data
processing. The loss of one of these factors would certainly
have an adverse effect on the value of the product.
The display of all of the data of interest is important
to the analyst for several reasons. Real time crop monitoring
is, to a large extent, an intelligence function. The data needed
are never complete and the quality of all of the data is usually
suspect to some degree. The key to a successful operation is the
ability to take all available, pertinent information and combine
it to produce the best possible answer in the time allotted. If
parts of the data are removed from the system, it is quite
possible for errors of judgement to be made. If, for example,
only the vegetation index values are available, a region with
partial cloud cover may be mistaken for a loss of vegetation
vigor. If all of the spectral information is available, the
brightness and the patterns of brightness will allow the true
situation to be determined.
The display of the vegetation index on the saturation
axis of the IHS color coordinate system allows the human mind to
see crop vigor as an independent feature of the false color IR
image. Relative crop vigor is then unchanged by changes in
brightness caused by atmospheric interference or surface
phenomenon. Redder pixels are better pixels regardless of the
situation. A thin cirrus cloud can partially cover a portion of
a crop region making the healthy red field appear pink, but
within the cloud covered region the better fields will still
appear redder (more saturated). The human mind will be annoyed
by the presence of the cloud but not confused, and the analysis
of the health of the fields will be done correctly.
The way the information is displayed in a false color, IR
images makes the identification of surface features, clouds, and
water possible. The pattern recognition capabilities of the mind
are at their best when viewing regions with partial cloud cover
of different cloud types. The amorphous patterns caused by
various cloud types and cloud systems are easily recognized by
the mind and naturally sorted out as not part of the surface
scene. The patterns of fields, terrain, and water bodies, which
are also amorphous, are very different and easily recognized. In
the false color IR image, it is quite possible to do this type of
pattern recognition rapidly and effortlessly, and with very
little training.
The production of the image itself requires no image
processing other than the decoding of the input data and the
necessary hardware to produce a color image. Each pixel is
assigned colors which are based only on the incoming sensor data
23
for that pixel. This type of image is the least expensive image
to process because no time consuming, complicated mathematics are
necessary to produce it.
Shortcomings of the false color IR image
The shortcomings of the false color IR image are (1) the
use of three color axes to display two dimensional data, and (2)
a lack of perceptual color resolution. When the success of the
image in crop monitoring is considered, these shortcomings may
not be serious, but they do provide the possibility to improve
image quality, and therefore, the value of the image.
In the false color IR image, the scene brightness is
displayed as intensity, the absolute value of the vegetation
index is displayed as saturation, and hue is used to display the
sign of the data: red for positive, and cyan for negative. There
is no color axis available to display such important information
as thermal IR. If thermal IR is substituted for either the blue
or the green axis of the RGB color coordinate system, the status
of the vegetation is no longer displayed in the minds' natural
coordinate system and one of the main positive attributes of the
image is lost. The result is a confusing image of little value
for crop monitoring analysis.
The choice of saturation for display of the vegetation
index is fortuitus in that it did produce an image which allowed
analysis to be performed. It is unfortunate in that it may not
be the best image which can be designed and it is not the
industry standard. The image was not designed for the purpose
for which it is being used and the task of designing a better one
will certainly involve changing the axes of the color coordinate
system. Of the three axes of IHS, the axis of saturation has the
poorest resolution. The chromaticity diagram compares saturation
and hue, and it is obvious that hue is the better of the two
axes.
The ACCS (Ambroziac Color Coordinate System)
The maximum color resolution possible for saturation
depends on the hue. Blue is the best, with 103.0 jpcd between
blue and white, while yellow is the worst, being only 26.7 jpcd
distant from white. The greater the path length in color space,
the greater the perceptual resolution along the path. Hue would
be a better choice because resolutions are better between hues
and the length on the line is unbounded.
The shape of the IHS axis is difficult to discuss in the
context of distance comparisons because it is 6-dimensional and
the distance calculations in this 6-dimensional space are not
worth the effort unless the intensity axis of IHS is being
considered to display the vegetation index as a replacement for
saturation axis used in false color IR. Intensity is certainly a
24
possibility for display of the index but the transformed image is
difficult to learn to use. People are comfortable with intensity
showing scene brightness even if it is brightness of an invisible
part of the spectrum. Bright colors representing high
reflectances requires no learning.
In the conical coordinate system, the false color IR
image color coordinate system is the plane containing red, white,
cyan, and black. Infrared reflectance is plotted along the line
black to red and visible is plotted from black to cyan. The path
followed in moving from red to cyan is not a geodesic path but
rather it is the path specified by the color changes on the CRT.
These paths are straight lines in Euclidean space and the
distances along some of the key paths are given in table 1. The
total color distance from red to cyan, the axis of the vegetation
index, along the path followed by the CRT is 78.2 jpcd (red to
white) plus 49.7 jpcd (white to cyan) for a total for 127.9 jpcd.
There are an infinite number of longer paths which could
be used to increase perceptual visual resolution, but care must
be taken that the increase in resolution is not offset by a loss
in some other attribute of the false color IR system. One way to
increase perceptual resolution is to choose 128 colors which are
as far apart as possible and order them to maximize the
differences between adjacent colors. The gain in resolution
would not be as great as the loss in information due to the
removal of false color IR's second attribute -- display of the
vegetation index on an axis of the minds natural coordinate
system. If more red is more biomass, the mind can see increasing
biomass under a wide variety of adverse conditions, but if the
order of the colors is not natural to the mind, it will lose all
analytical ability, for if relative change is not displayed then
relative analysis is not possible.
If hue is used in place of saturation, the possible color
length of the vegetation index axis is increased from 127.9 to
450.3 jpcd, which is a 3.5 fold increase in perceptual color
resolution over red and cyan saturation changes, and no image
attributes are lost. It is not practical to use all of the
possible hues in the new coordinate system, nor does the system
need to cover all possible combinations for visible and near IR
sensor readings.
One of the most difficult tasks for the scientist working
with color is to create a hue scale that looks reasonable. Such
a scale has been calculated for 100 hues and 256 hues. Since hue
scales are display devices and viewer dependant, some
modification is almost always needed, but a good starting point
is nice to have.
The Ambroziak Color Coordinate System (ACCS) image color
coordinate system, which was chosen for testing, was one which
replaces the positive vegetation indexes with changing hue and
gives all negative index values a single hue which contrasts with
the positive NVI hues. The result is an image with (1) all of
25
the positive attributes of the false color IR image, (2) 3.5
times the perceptual resolution of false color IR, and (3) which
uses only two coordinate axes leaving one axis free for
additional information such as thermal IR data.
The new coordinate system used equally spaced hues from
red through green to blue to display the positive values on the
NVI. There is no loss of information for analysis because all
possible values of naturally occurring combinations of visible
and near IR reflectance are assigned a unique color. The
negative vegetation indexes are not as important as the positive
in crop status assessment and all combinations of visible and
near IR reflectance are not possible in the negative index region
of the reflectance graph. The negative index values are
basically one dimensional in the visible and near IR portion of
the spectrum. Pure, deep water is near the origin and moves in
an arc towards the zero NVI of whatever substance is beneath the
water or suspended in it as the depth decreases or turbidity
increases. These positions are identified well enough by
intensity alone for crop assessment analysis.
The region of the ACCS covered by water has at least the
same perceptual resolution as the false color IR image. In the
false color IR image, the water is shown by blue and/or cyan.
Depending on the satellite used to create the image deep, clear
water is either very dark blue (Landsat) or cyan (AVHRR). In
both cases, the color is black on most images and hue has no
meaning. As the water becomes more shallow or more turbid, the
intensity increases in false color IR and in the ACCS images,
both producing similar results. The major difference in the ACCS
is that the saturation does not decrease as it does in the false
color IR image. The ACCS displays water as magenta in all cases.
The use of hue instead of saturation does not violate the
use of the minds' natural coordinate system to display the
normalized vegetation index since hue is one of the axes of the
IHS system. The use of hue for both the sign of the index and
its magnitude is quite valid because both the index and the hue
are represented by angles. The normalized vegetation index is an
angle measured at the origin. Hue is an angle in the hue
saturation plane of the IHS conical type coordinated system. The
correspondence is at least as natural as the use of saturation
from index values of minus one to plus one. The false color IR
image coordinate system slices though the IGS coordinate system
cone while the ACCS wraps around the outside.
The coordinate system makes a distinction between land
and water which is greater than that made by the false color IR
image. The color of soil on the false color IR image is grey or
white and anything but black water is unsaturated cyan. The
maximum color distance from cyan to white (assuming cyan of
saturation = 1) is 49.7 jpcd while the distance form all water on
the ACCS to soil is the color distance from magenta to red which
is 59.2 jpcd (table 1). The color distance along the intensity
axis is the same in both coordinate systems. The patterns of
26
land and water are, therefore, as visible or more visible than
those patterns on false color IR.
An annoying feature of the ACCS is that clouds are bright
red rather than a natural white as they appear on false color IR.
This can be overcome by making the pixels with high visible
reflection values white, because clouds are bright in the visible
and nothing else except bright sand is that bright. A better
solution is to use the thermal channel to detect clouds and make
them white. Temperature of the cloud can be displayed by
saturation changes so that the saturation channel is used to
display temperature making the ACCS a true three channel display
with white clouds.
Using this IHS color coordinate system, it was possible
to do analysis of African vegetation which could not be done with
false color IR images. In training courses, repeated attempts by
students to use the false color IR images always resulted in
failure while using the ACCS was usually successful. Color
vision problems of some students prevented one color coordinate
system from being universally better that the others, but the
best analysis was always done using the ACCS.
SUMMARY
Each application of color to an analysis problem is
unique and no single color system can be called the best. Since
color is more psychological than physical, some trial and error
will always be a part of any color coordinate design process.
The best tool to have for this process is one which is as
versatile, as well as general and as controllable as possible and
make sure you can change it, because no system design will ever
let you do everything you will want to do. To use these systems,
a working knowledge of color theory is a must and the more you
use it, the easier it will become to use.
I believe that image processing is the evolutionary
process of putting any color of pixel anywhere you want to, and
every time you do it, you think of a better way to do it. There
are so many variables that building a truly generic system is
impossible at best.
27
