The Colors of the Stars

Thor Olson
Management Graphics Inc.
Minneapolis, MN

100:1, not a coincidental match to the human visual range

Abstract for a given adaptation. This is a severe restriction; of the
millions of stars we might want to show, only a few hundred
The brightness and colors of the stars have fascinated could be accommodated.
observers of the night sky for millennia. It is difficult
however, to portray the depth and amount of visual It would be nice if star images could also convey
information of the heavens on actual displays. This paper something about the nature of their light. Although most
applies principles of reproduction, color science, and stars appear to us as bluish white, the stars vary enormously
perception, to generate attractive and information-rich in temperature, and consequently, color. Presenting stars in
starfield images. their correct color would provide a natural indicator of a
correlated physical characteristic.
Although the casual appearance of stars is bluish-white,
starlight actually ranges along the entire black-body locus. Although an exact duplicate of the night sky may not be
Scotopic (night) vision cannot see the full color that would practical or even useful, this is actually a rather familiar
result, but we can use color as a natural way to depict the problem in the area of graphic arts and imaging. It is almost
various types of stars and their surface temperature. This is always the case that we do not want to duplicate a scene, but
done in three steps. First, a relation between the instead want to duplicate the appearance of that scene. We
astronomical color index of a star (the difference in power apply principles of reproduction, color science and
between two historically established spectral bands) and the psychophysics to accomplish this. This paper describes the
star’s effective surface temperature is obtained. From this application of those principles to make information-rich and
black body temperature, colorimetric coordinates are natural-appearing, even beautiful, representations of the
calculated. Finally, the colorimetric values are converted to night sky.
device coordinates for the display.

Historical notes on star color and magnitude

Introduction and motivation measurements
Making pictures of the night sky is an ancient activity. Before the invention of photography, stars were charted
Apart from astrological efforts to foretell the future, more according to their relative visual brightness. Astronomers
successful uses for mapping the sky have been found for became skilled at making judgments of brightness by
calendars, timekeeping, navigation, and deepening our comparisons with reference stars. A scale of measurements
understanding of the universe. And beyond these practical evolved where the brightest stars, those “of the first
applications, star-filled skies find their way into the esthetic magnitude” were judged to be twice as bright as second
and spiritual domains of art and literature. magnitude stars, which were twice as bright as third
magnitude stars and so on. The limit of naked-eye
It would be useful for educational and reference astronomy was around magnitude 6. Much later, when
purposes to make maps of the sky that portray the stars in an instruments could measure actual stellar luminances, it was
accurate way, their positions and relative brightness discovered that a difference of 5 magnitudes was not 32:1 as
correctly displayed. Making such a technically accurate expected, but was instead 100:1. This means that a stellar
image is not possible with common display technologies magnitude unit is equivalent to the 5th root of 100, or a
however, and even if it were, its usefulness would be limited factor of about 2.5. (This historic anecdote provides a clue
to about the equivalent of walking outside at night and regarding the visual brightness response of scotopic vision).
looking up. The dynamic range of naked-eye stars is about
 When photographic films became sensitive enough, The spectral differences that led to B and V magnitude
they were put to the task of automating the measurements of measurements was formalized by astronomers. Using
star magnitudes. It became immediately apparent however technology available at the time, specialized glass filters
that the magnitudes as recorded on film did not match the were selected that approximated the spectral sensitivities of
magnitudes that had been so tediously obtained by observers photographic film and human visual response. Plots of them
over the years. The discrepancy was traced down to the are shown in figure 2. If you were to make measurements of
difference in spectral sensitivity between the film (which in stars through these filters you would find that their ratio (or
those days was blue sensitive) and the scotopic sensitivity of difference in magnitude units), provides an indirect measure
humans. of their temperature. The use of “magnitude” as a unit was
preserved, and even though these are actually radiometric
Even though there was not a perfect correlation, making measurements, astronomers refer to this activity as stellar
magnitude measurements with film was so vastly more photometry. The difference between the B and V
efficient that stars were catalogued according to their “B” magnitude measurements became known as the stars color
(blue) magnitudes in addition to their “V” (visual) index.
magnitude. It was clear that B and V were samples of a
continuous spectrum from each star, and the difference B-V
became known as the star’s color index.
6
U
Converting color index to black body B
5
temperatures
V
The spectra of the sun and stars is very complex, with
Relative sensitivity
4
emission and absorption lines that indicates the chemistry
and physics going on at their surfaces and in their
atmospheres. But underlying the complexity is a continuous 3
spectrum that can be approximated as an idealized black
body radiating at the star’s characteristic temperature. The
0
temperature of stars ranges from 3 K (if you include cold 2
0
dead stars) to over 50,000 K. The spectral power radiated
by them follows laws discovered by Max Planck at the
beginning of the century. A sampling of the relative spectral 1
power output for black bodies at various temperatures is
shown in figure 1.
0
250 300 350 400 450 500 550 600 650 700 750
Power spectra of black body radiators
10000 wavelength, nm
25000K
1000 Figure 2. Spectral characteristics of the U, B and V bands used in
20000K
astronomy.
100 15000K
10000K
10 An exact formula relating the color index, B-V, to the
8000K
effective black body temperature is not known to this author,
1 7000K
but the energies transmitted through the B and V filters can
6000K be computed, and effective temperatures plotted for various
0.1
5000K luminosity classes of star spectra (figure 3). An
0.01 approximating formula was obtained to fit this data. From it
4000K
0.001 3000K one can find the effective temperature of a star given its
color index.
2000K
0.0001
5000
400 450 500 550 600 650 700
Teff ≈ 1000 + (1)
wavelength (nm)
(B − V + 0.5)
Figure 1 Spectral power output for black body radiators at various
temperatures.
 30000 Converting color to device coordinates
● ● Luminosity Class V
Effective Color Temperature

25000 We are now faced with how to present the color on a

◆ Luminosity Class III
real output device. Up to this point the conversion from
20000 ■ Luminosity Class I stellar magnitude measurements has followed well defined
1000 + 5000/(B-V+0.5) operations. We will now have to make some judgments,
● guided by principles of perception, on how best to present
15000
this information on a physical device. Two types of output
10000 ● will be discussed: a CRT monitor, and film transparencies.
●
● ●
◆
■ ●●
◆ ■●◆
5000 ◆■ ● ●■ ◆ CRT displays
■ ◆■
◆●
A CRT is a well behaved color device. Once it has
0 been calibrated and characterized, the colors it makes are
-0.5 1 0 1.5 0.5
2 predictable and easy to obtain. Say that a monitor has been
B-V set up so that its white level matches the D65 daylight
Figure 3. Effective black body temperature as a function of color whitepoint. If it is using the standardized phosphor set in the
magnitude index B-V CCIR Rec. 709 video standard, then the linear RGB levels
can be obtained from an XYZ triple by the following matrix
Converting temperature to color operation:

Just as we can compute the energy through the B and V R   3.240 −1.537 −0.499  X 
filters, we can obtain the energies sensed by the CIE     
G =  −0.969 1.876 0.042  Y 
colorimetric standard observer and obtain chromaticity (2)
coordinates for various black body temperatures. In fact, B   0.056 −0.204 1.057  Z 
this was done long ago and the locus of black body radiators
is shown on many chromaticity charts as a reference (figure
4). A polynomial fit to the locus is provided in Hunt [1] The actual CRT drive levels R’G’B’, are obtained from
which can be used to obtain the chromaticity of a star from RGB by applying the power law of the device which, when
its temperature. calibrated to the video standard, uses a gamma of 0.45
(approximately 1/2.2).
0.9
R’  R 
0.45
0.8
   0.45 
G’ = G 
  
(3)
0.7
B’  B 0.45 
0.6

0.5 3000 K One may now take the chromaticities along the black
y 5000 K body locus representing stars at those temperatures, scale
0.4 • • • 2000 K them to obtain XYZ tristimulus numbers with Y=1 (in other
•
•
••
words, treat it as if it was an adaptation whitepoint), and
0.3 ••• calculate video R’G’B’. This has been done for a collection
10000 K of temperature points along the locus and is plotted in figure
0.2 5. This plot tells us what monitor R’G’B’ levels to use to
display stars at a given temperature.
0.1

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
x
Figure 4. Locus of black body radiators on an xy chromaticity
diagram.
 2 will keep all of the temperatures within range, and they will
be chromaticity-correct, but the relative perceived brightness
1.8 of differently colored stars will be skewed. Low
1.6 temperature red stars especially will seem dimmer than they
should (remember, our sample star colors were scaled so
1.4 that their luminance level, Y, started out the same). From an
esthetic viewpoint, suffering this brightness error is
Video level

1.2 preferable to an overall dim picture. Fortunately there is a

1 way to restore the brightness, described later in the treatment
of actual star magnitudes.
0.8

0.6 R’709 Film images

Compared to CRTs, film is considerably less well-
0.4
G’709 behaved. Although it has a large dynamic range and color
0.2 gamut, there are variations in film batches and processing
B’709 chemistry. On top of this are differences in projector bulbs
0 and room lighting. Understanding the colorimetry of
1000

2000

3000

5000

6000

7000

8000
4000

9000

10000
projected transparencies has been tackled by several
0

researchers who continue to discover its complexities

[2][3][4]. Nevertheless, because of the drama and depth it
temperature, Kelvin
adds, and the natural similarity of a darkened theater to a
Figure 5. Video R’G’B’ to represent star color as a function of nighttime environment, it is a prime output medium for this
temperature. project. If we are willing to sweep the uncontrolled
variations under the observer’s adaptation rug, we can still
There are some interesting things to point out in this obtain perceptually satisfying images of the stars.
plot. First, as astronomers know, there are no green stars!
At no temperature does one find that the energy is highest in The range of star colors is not so high as to strain the
the green band. Either red is dominant or blue is dominant. limits of color slide film. In fact, they are relatively
Even at the point where they are equal, they are both greater unsaturated colors, and because of this, the linear matrix
than the contribution from green. type operation is successful. The matrix for Ektachrome
film exposed on a Management Graphics Solitaire film
The green channel stays relatively constant. By the recorder is:
selection of our device whitepoint (D65) green equals unity
R   3.275 −1.557 −0.511 X 
0
at or near 6500 K. At that point red and blue are both
greater than one. What this means is that the black body     
G =  −0.781 1.693 0.045  Y 
locus does not pass exactly through the D65 whitepoint (if it (4)
0
did, all three components would equal 1 at 6500 K). B   −0.030 −0.119 1.054  Z 
There is another practical matter to consider. Figure 5
shows the monitor levels normalized to 1, usually taken to Film recorders are typically calibrated to a power law
be the maximum drive level in each color channel. In order with a gamma of 2.2, so there is the same nonlinear
to accommodate the full temperature range, we must scale transform to R’G’B’ as for CRT monitors.
the levels so that no color channel will run out of range. The
scaling must be done in RGB prior to its nonlinear Measurements were made on some test “stars” made
conversion to monitor R’G’B’. A factor of 2 would be from the formulas above and imaged onto Ektachrome film.
0
enough to display stars down to nearly 3000 K or so. But The samples are illustrated in figure 12 and their
0
this would cause the near-white stars at 6500 K to be chromaticity measurements plotted in figure 6. The success
displayed at a rather dim level, considerably down from in the match with the targets indicates that film can be
white. treated as a linear system over the small range that these
black bodies cover.
There are several approaches to solving this. One is to
simply accept the dim white point, knowing that the relative
color strengths are correct. We could also scale the levels so
that the maximum of each RGB set was normalized. This
 0.6

0.5

2000 K
• • •∆
y chromaticity

• ∆
∆
0.4

∆∆•∆
• •
ƥ
0.3
••
∆
10000 K

0.2 • target

∆ measured Figure 7. A section from a Tirion star map.

0.1
0.2 0.3 0.4 0.5 0.6 0.7
Before electronic photometers became available, and
x chromaticity
even now when making large surveys, star magnitudes were
obtained by measuring the diameter of their recorded image
Figure 6. Measured colors from Ektachrome on film. The principle behind this is shown in figure 8. The
star’s image is not a point, but is the point spread function of
the lens viewing it. This function will depend on the
diffraction limit of the aperture, and the aberrations of the
Perceptual mapping of stellar magnitudes optics, but usually has most of the energy localized at the
center and then trailing “skirts” farther away. The behavior
So far we have discussed only the color of stars. The
of film is to convert to metallic silver those areas that are
other major characteristic of the night sky we would like to
exposed with enough energy to exceed a certain threshold
portray is the enormous range in their apparent brightness.
level. Because of this, brighter stars, having a higher
The naked eye can see stars down to magnitude 6 (on a clear
amplitude point spread function, will make a larger recorded
dark night). But the sky is filled with stars dimmer than this
spot. The relationship is not proportional, but depends on
that show up when you look closely with binoculars and
the details of the point spread function of the system, and the
telescopes. Can these be shown in a meaningful way in a
threshold of the film.
star field image? Further, and more basic, stars are
essentially point sources. How does one represent a point
The human eye does not behave like film, but does
source on a output device that uses finite-sized pixels?
share some characteristics. The diffraction and aberrations
of the lens produces an optical point spread function, and the
As mentioned in opening, maps of the sky have been
receptors in the retina have thresholds that cause neuron
made for a long time. Modern star maps have settled on
activity. Unfortunately, there is a great degree of retinal
various conventions to display the brightness and types of
processing that prevents a simple relation between
stars. A look at one of the maps of sky cartographer Wil
luminance and perceived brightness. Much research has
Tirion [5] shows the common practice of using larger
been conducted into this topic, but it is largely limited to
diameters to represent brighter stars. For a black and white
experiments using area patches of known luminance, not
chart, this is really the only option available. It has a basis
point sources and their equivalent area stimuli.
however, in how stars are recorded photographically, and
are perceived visually.
 1

threshold
0 Figure 9. An unrecognizable constellation Leo, rendered using the
-5 5 equal energy hypothesis.

This worked well for bright stars which had visibly

distinguishable areas, but it caused difficulties in
discriminating the dim stars that were imaged using very
small areas. It seems that once the retinal image area
approaches the eye’s point spread function (where small
areas look the same as point sources) one needs to return to
the principle of delivering similar energies in order to
Figure 8. Recorded spot sizes on film for various intensities (a achieve similar brightness perceptions.
four magnitude range) in a diffraction limited system.
A single formula that behaves in the desired way at the
The equal energy hypothesis extremes, and makes a nice transition is:
In lieu of the missing psychophysical data, some
experimentation was done. We would like to obtain a scale L
R2 ∝
[ ]
of small areas that produce that same relative sensations as 2 (5)
their corresponding point sources. An early hypothesis L 3 + LA
translated the intensity of a star directly into an area on the
image. This would preserve relative energy, a magnitude 5
star would have 2.5 times the area of a magnitude 6 star. where R is the radius of the displayed star, and L is the
But making the area of a star image proportional to its relative luminance of the star obtained from its visual
luminance was disastrous. The range in star sizes became magnitude M:
enormous, obviously not corresponding to a perception of
the night sky (figure 9). −M
L = 100 5
(6)
The equal sensation hypothesis
Making the area proportional to perceived brightness on
the other hand was quite believable. Brightness is LA is a transition luminance, corresponding to a
asymptotic to the 3rd root of luminance, so the growth in transition magnitude where the area of the displayed star
1/3
area was considerably smaller, √2.5 or 1.36 per magnitude
3 changes from being proportional to brightness (L ) for
bright stars, to being proportional to luminance, for dim
unit. This strategy makes some sense. Consider a point
stars. The selection of this transition point will depend on
source that is 1 magnitude (2.5x) more luminous than
the visual angles covered by the star images.
another. If the intensity of the sources are beyond the
adaptation of the eye, then according to the CIE functions
This produces the relation plotted in figure 10 where the
[6], it will be perceived as 1.36 times brighter. If one wants
transition magnitude was M=2. This plot also shows the
to produce the same total sensation as this, but with a fixed
scale used in Tirion’s star map. It is not known how he
luminance within the eye’s current adaptation, an area that is
devised the scale, but it is satisfying that it is similar to our
1.36 times larger seems reasonable.
heuristic derivation.
 10
Area=B Recovering the displayed color range
Ma=2
The star magnitude to size and amplitude relations can
now be put to use to solve the problem of displaying the full
Area=L color range of the stars. The difficulty was that to show the
1 • • • •
correct colorimetric range, we had to accept either a
• •
displayed radius

Tirion brightness error, or a dim white point. We can now restore

• the whitepoint to its normal full level, normalize star
• amplitudes for their color, and adjust the displayed size of
the star to preserve relative brightness. For example, a low
• temperature red star would be amplitude normalized to its
0.1 red channel, reducing the green and blue. This reduces its
visual magnitude (brightness). The perceived brightness is
restored by increasing the area used to display that star.

In effect, we have made a tradeoff. We are obtaining an

improvement in the perceptual presentation of the stars, but
0.01
the calibration between magnitude and displayed area now
-2 0 2 4 6 8 10
depends also on the color. The resulting color scale is
Star Magnitude
shown in figure 12, where the adjustments in size can be
Figure 10. Relations used to draw stars using area as an observed to preserve an overall impression of similar
indication of their magnitude. brightness.

The limits of area equivalence

In continuous tone images, we have the luxury of being Additional elements in starfield illustrations
able to reduce the intensity of the star in addition to
controlling its rendered size. Over a large range of star Almost by definition, a good map is an effective
magnitudes, the relative area between the brightest and presentation of information. A poor map may contain the
dimmest star becomes quite large. If the dimmest star is same data but presents it in a way that makes it difficult to
rendered using a single pixel, it is possible that the brightest understand. The choice of scales, symbols, colors, and fonts
star could become excessively big, consuming too large a are critical to the success of the map. There are now a
fraction of the display area. To solve this, the brightest star number of “desktop planetarium” programs, and most will
is set to an “esthetically acceptable” diameter (this judgment make star charts of a selected area of the sky. In some, the
will depend on the purpose and resolution of the image). All user is free to make color choices for symbols, coordinate
dimmer stars are then smaller than this. Eventually a star grids, and backgrounds. It is observed by this author
magnitude will be encountered which would require less however, that any choices much different than the traditional
than 1 pixel. We can now reduce the amplitude of the pixel colors tend to make it more difficult to read the chart.
to preserve the brightness relation. Instead of the amplitude
being fixed and the size reduced, we now convert to keeping There is probably a large body of cartographic
the size fixed (at more than 1 pixel if we desire) and knowledge that helps a professional mapmaker to make
reducing the amplitude. We can continue this until we have these color choices. It is interesting that the most effective
reach the dimmest level of the display device, at which point map is also an attractive map. There are likely esthetic and
we have reached the limit of stellar magnitudes that we can psychophysical reasons for this, but it is beyond the scope of
display in their proper relative brightness. this paper to explore that direction.

The reduction in amplitude for each magnitude should be the Additional elements that make a rich map of the sky
same as the reduction in area. Since this presumably occurs include gridlines showing the celestial coordinates, an
for small dim stars where the area was proportional to indication of the faint band across the sky made by the
luminance, a factor of 1/2.5 = 0.4 is needed. Since the Milky Way, and sometimes, construction lines showing the
image display follows a power law, the reduction in drive constellations.
level for each magnitude step is only 0.66. The
combination of using area and amplitude was used to make A natural background for an image of the night sky is
the magnitude scale shown in figure 11 (which may not black of course. Although difficult to print, it makes the
reproduce to its full range in this publication). stars stand out in excellent contrast in a transparency. The
selection of gridline colors is made so that the grid is an
unobtrusive element, almost invisible, until one wants to see
it, at which point is should be visible for unambiguous References
position reference.
1. R.W.G. Hunt, Measuring Colour, Ellis Horwood
The Milky Way is an awe-inspiring feature of the night Limited, West Sussex England, 1987.
sky. The faint glow of the band of our galaxy is not really
seen, but rather sensed by the low light sensitivity of off-axis 2. Robert Hall Wallis, Film Recording of Digital Color
rod cells. One has the feeling that it is a bluish white glow. Images, USCIPI Report 570, May 1975.
This is due to the high percentage of young hot stars (blue
stars) in the nearby arms of our galaxy. Rendering the 3. Mark Fairchild, Roy Berns, Audrey Lester, and Hae
Milky Way at the infinite temperature limit for black body Kyung Shin, Accurate Color Reproduction of CRT-
Displayed Images as Projected 35mm Slides, IS&T/SID
radiators provides one more color reference, albeit
Color Imaging Conference, Scottsdale AZ, 1994.
somewhat artificial, for making a visually appealing map.
4. Audrey Lester and Mark Fairchild, Thermochromism of
Constellation lines are a dilemma. Like the coordinate Ektachrome 100 Plus Professional Transparencies Upon
grid, one wants them to be available to help locate these Projection, Journal of Imaging Science and Technology,
major star patterns. But once found, it would be nice if they v.38, n.4, July/Aug 1994.
could disappear completely, leaving the beauty of the
starfield to make its own patterns in the mind of the viewer. 5. Wil Tirion, Cambridge Star Atlas 2000.0, Cambridge
If one includes the constellation lines in an illustration, they University Press, Great Britain, 1991.
can be rendered in a faint, transparent color, say greenish,
that contrasts with the star colors and the Milky Way. 6. H.W. Bodmann, P. Haubner, A.M. Marsden, A Unified
Relationship between Brightness and Luminance, CIE
Proceedings 19th Session, Kyoto Japan, 1979.
Combining all of these elements into the color and
brightness-scaled star fields yields the typical view shown in
7. Elske v.P. Smith and Kenneth C Jacobs, Introductory
figure 13, a satisfying and informational image. Astronomy and Astrophysics, W.B. Saunders Company,
Philadelphia PA, 1973.
Beyond the flat map
There is another dimension that can be utilized to
8. Tom N Cornsweet, Visual Perception, Harcourt Brace
present information from an image: depth. A stereo view of Jovanovich, Inc.Orlando FL, 1970.
the sky would normally be perceived as the sky looks: flat.
The stars are just too far away to provide any significant 9. David H Levy, Skywatching, The Nature Company,
parallax. But there is no reason we cannot amplify the small Berkely CA, 1994.
parallaxes that have been measured by astronomers to
illustrate the relative distances of the stars we see. This has 10. David Malin and Paul Murdin, Colours of the Stars,
been done in figure 14. If one can fuse the two images in Cambridge University Press, London, 1984.
the pair, either through free-viewing or using a stereo
viewer, one can get a glimpse of how God might see this 11. M.J.G. Minnaert, Light and Color in the Outdoors,
Spreinger-Verlag, New York, 1974.
part of the sky (provided God has trichromatic night vision
and an interocular distance of 1.6 light years)!

More questions than answers

As usual in investigations of Nature, as questions are
answered, more are posed. If we can display the colors of
0
stars, and a 6500 K star looks white, why does the Sun (a
0
6800 K star) look yellow? And why is the sky blue, really?
Shouldn’t the sky scatter starlight as it does sunlight? Why
don’t stars look red when they set? The images created
here are not calibrated views of the sky. Nevertheless, for
educational and artistic purposes, it seems that it is
worthwhile to pay attention to the principles of esthetics,
color, and perceptual sciences.
 Figure 11. Rendered magnitude scale using both area and intensity to extend its range.

Figure 12. Rendered color scale using area variations to preserve brightness at different temperatures.

Figure 13. A star map showing Taurus and Orion using the color and brightness relations described in this paper.
 Figure 14. A stereo pair of star images showing the constellation Leo “as God might see it” (with full color vision and
interocular distance of 1.6 light years). Free-view them or fuse the images using a stereo viewer. If free-viewed with crossed
eyes, the distances will be inverted: the bright constellation will appear to be behind a foreground curtain of faint stars.