Color dictionaries are created for several purposes:
-
Standardized color names facilitate specification, purchase, and use of
colored goods, markers, etc.
-
Vendors like Pantone and Toyo register colors (and names), providing
formulas for inks, plastics, toners, and paints to guarantee uniformity
and accuracy of color for their clients' products. These color
designations are often just numbers and letters; requiring search
through process guides to find a particular shade.
-
The combinatorial color dictionaries underlying the
Munsell, OSA-UCS, and GIA color scales allow field-workers to encode
and communicate color from visual observation. These systems endeavor
to partition their color spaces into equally distinguishable regions
with a named color at the center of each.
-
A set of color names can be used to restrict selection when a spectrum
of colors is not available.
-
On computers, summoning colors by common names relieves the tedium of
adjusting or mousing each color used.
Color dictionaries for field work must be small enough to make
reasonably quick determinations. A couple hundred names seems to be
an upper bound.
Larger dictionaries which range into the thousands were created for
identifying paints or inks. Common color names (like "blue") are not
used alone; but they can be components of names. Such collections of
names are meaningless without their charts or samples. These are
referred to as idiosyncratic in this article.
Color dictionaries should provide a means for specifying colors
independent of the
gamut limitations of
displays and printers, but all the online materials I found in 2001
used RGB coordinates or proprietary unconvertable denotations.
More problematical is the approach to color assignment taken by most
of the authors. One can either try to choose colors matching the
color names; or try to find names to describe the colors.
Unfortunately, most of the online color dictionaries use the latter
approach. Not only are most of the colors they assign located on an
RGB grid, they assign common names to the extreme points on the sRGB
cube. Most of these extremes lie outside of the print gamut.
The limitations of the print gamut result primarily from the physics
of reflecting color. Surfaces able to reflect RGB primaries at full
brightness when illuminated with the full spectrum (D65) light require
impossibly steep spectral responses. These extreme colors look
fluorescent or just wrong when used in images.
Although not completely free of limitations, pure primary light
sources are more practical. Thus when building scenes, the color
dictionary for lighting and glowing objects should be different from
the color dictionary for surface colors.
The Catalog column in the table below links to PDFs displaying
triangles, each filled with a color and its sRGB coordinates and name
below it. Because the sRGB coding has limitations, those colors not
inside the sRGB gamut are marked with an X. The width of the
X varies as the number of sRGB primaries not within bounds.
When a swatch (other than white) is marked with a X, you can't
be sure you are seeing the intended color.
Color Space Dimension Reduction describes the
mathematics used to produce these usable color catalogs (unlike the
one to right).
Under the sRGB Space column are PDFs showing the RGB color cube
in perspective with a colored ball for each color in the dictionary
placed within the cube.
The L*a*b* Space column links to VRML-2.0 files showing a
sphere for each color floated in CIEL*a*b* space. Some screen shots
of L*a*b*-space are shown in the Resene and
NBS/ISCC sections.
Web-searches turn up few original color-name dictionaries. The most
replicated one is `/usr/X11R6/lib/X11/rgb.txt' from the X11
window system distribution. Last modified in 1994, does its vintage
indicate perfect wisdom of named colors?
- Color Name Dictionary: X11 rgb.txt
- Aside from its numbered grays and greys, `rgb.txt' has
54 duplicates out of 455 colors. The
duplications are not entirely consistent: slategray1, ... slategray4
are present; but slategrey1, ... slategrey4 are missing.
200 of its color names are devoted to `gray0' through
`gray100' and `grey0' through
`grey100'. The intensity values given to each
grey<k> is k * 2.55. These grey values are not equally
spaced in perceived brightness. And they would only produce equally
spaced photometer readings when displayed on a linear RGB monitor.
Although the default X color model (RGBi) is linear, a
catalog of X11 numbered-greys shows that
linear RGB709 biases the perceived `rgb.txt' colors too
light. They are more believable when treated as sRGB values. But
even then, the bulk of the colors are still too light.
When viewed in sRGB space, the grid nature of
`rgb.txt' is evident; as is the relative emptiness of the
darkest octant.
When viewed in CIELAB space, the
`rgb.txt' color set is top heavy, having few dark tones and
uneven spacing.
In summary, X11's `rgb.txt' is suitable for light sources,
but poorly suited for surface colors; and should be interpreted as
sRGB in either case.
As detailed by
Xorg Bugzilla Bug 12293
Cannot install custom rgb.txt file in RHEL 5,
Xorg version 7.1 completely ignores rgb.txt
(still present at /usr/share/X11/rgb.txt).
This bug was submitted 2007-09-05, and has had no activity as of
2008-02. There appears to be no way to install a better color
dictionary without modifying and rebuilding Xorg Version 7.1.
- Color Name Dictionary: X11 Xcms.txt
- In 1992, the Xlib extension "X Color
Management System" was completed.
But `xc/lib/X11/Xcms.txt' in the X11
source tree is not installed by XFree86, and contains only 20 colors,
half of them numbered grays:
This device-independent color name database is provided just as an
example. There are no claims that these color specifications are or
should be bound to the corresponding color name.
- Color Name Dictionary: X11 old-rgb.txt
-
`xc/programs/rgb/others/old-rgb.txt'
in the X11 source tree is not installed by XFree86. All but 2 of its
72 color names are contained in
`rgb.txt', but most of its shades differ.
xc/programs/rgb/others/README says:
"old-rgb.txt" is the version that was shipped in previous releases; this
was originally "tuned" for the Digital VT240 series terminals.
- Color Name Dictionary: X11 raveling.txt
-
`xc/programs/rgb/others/raveling.txt'
in the X11 source tree is not installed by XFree86. It is a near
ancestor of `rgb.txt', only 32 of its
447 colors are not contained in
`rgb.txt'. A dozen of the remaining shades differ. I find
nothing to recommend `raveling.txt' over `rgb.txt';
they both have the problems mentioned in the section about
`rgb.txt'.
xc/programs/rgb/others/README says:
"raveling.txt" has lots of new colors, tuned by Paul Raveling at ISI for
the HP monitor.
Notes from Paul Raveling:
-
Many colors have been tuned for an HP monitor -- mine, to be exact.
Some of the old values were obnoxious enough to bring complaints from
users (like "That's Wheat???!!!"); so far early user reports on the new
RGB database are favorable.
-
File rgb.txt was reorganized into 3 sections:
-
Light and off-white colors, copied from several
Sinclair Paints color samples. The intent for adding these is to
provide a better choice for light-colored window backgrounds.
-
BTW, I wanted to find ANSI standard colors, but ANSI happily gobbled my
$16 without sending the specification I ordered, Then they ignored my
followup letter. Nuts to ANSI & "ANSI standards".
-
Special colors such as black, white, and favorite shades of gray.
-
A spectrum of colors, arranged to transition gradually between nearby
colors, running from generally blue colors through green and ending with
generally red colors. This includes all colors from the old X11R3
database, but they're no longer in (mostly) alphabetic order.
-
The gray scale from the original X11R3 database.
Within the "spectrum of colors" section there are clusters of colors,
each consisting of:
-
One or more Specially named colors. If more than one is present, all
are related by lying on a common line running from RGB = (0 0 0) [pure
black] to a single point on the surface of the RGB color cube.
-
Four colors at particular points on the same line in RGB space. Their
names end in "1", "2", "3", and "4", with "color1" being at the surface
of the color cube and the others at increasing distances approaching
black. Distance of these color points from black is approximately
logarithmic. This attempts a rough fit to human perception's
sensitivity to intensity.
Here's an example of one of these clusters:
210 105 30 chocolate
139 69 19 saddle brown
139 69 19 SaddleBrown
255 127 36 Chocolate1
238 118 33 Chocolate2
205 102 29 Chocolate3
139 69 19 Chocolate4
Note that the "original" colors, in this case chocolate
and saddle brown, don't always match the scaled points.
Does anyone think these changes are a good idea? Crummy idea? Do you
have other favorite colors? Can anyone do better at matching some tough
colors? Some colors, especially reddish ones, were VERY hard to
reproduce, & I'd welcome contributions from other HP users who can get a
better match.
The clusters were a bad idea. A simple function could have
allowed any degree of graying without cluttering up the database.
- Color Name Dictionary: X11 thomas.txt
-
`xc/programs/rgb/others/thomas.txt' in
the X11 source tree is not installed by XFree86. All of its 71 color names except 'aquamarine' are contained
in `old-rgb.txt'. Nearly all its shades differ from both
`rgb.txt' and `old-rgb.txt'. It is a huge improvement
from `old-rgb.txt'.
xc/programs/rgb/others/README says:
"thomas.txt" is a version of the older database that was tuned by John
Thomas at Tektronix to match a box of Crayola crayons.
Notes from John Thomas:
Advised by our human factors folks that "standard" named colors exist,
but only for well-controlled color coordinate systems (like CIE, but not
for RGB), I sat down one evening with the handiest standard of
subjective color names, a box of 72 Crayola crayons. (Believe it or
not, over 50% of the colors from rgb.txt were represented.)
Using an X-client implementation of the TekColor model, I created the
following list of named colors. Appearance on your monitor may vary
because of brand, age, and video drive circuitry, but I think you will
find it a better match for the average monitor, than the original
rgb.txt file from MIT.
John C Thomas
Tektronix, Inc.
Wilsonville, OR
jct@windex.TEK.COM
It is clear from comparing Thomas's sRGB to
old-rgb's sRGB that only Thomas looked at
the crayons! Except for 10 saturated colors, `thomas.txt' is
close to realistic for its small set of surface colors.
Two4U's Color
Page offers 3 color dictionaries: big, medium, and small.
- Color Name Dictionary: Two4U's Big Color Database
-
`big-txt.html'
is an uncredited replica of X11's `rgb.txt' with the addition of
`light', `medium', and `pale' --
apparently transcription errors of `light goldenrod',
`medium aquamarine', and `pale violet red'.
- Color Name Dictionary: Two4U's Medium Color Database
- Containing 100 colors,
medium-txt.html
adds 28 colors to `old-rgb.txt'. Of the shared names, only
Orange and OrangeRed differ. Most of the 28 added colors are
reasonable, but the bulk of the inherited `old-rgb.txt' colors
are not.
- Color Name Dictionary: Two4U's Small Color Database
- Containing 47 colors,
small-txt.html
is a subset of `medium'.
- Color Name Dictionary: U. of N. Iowa List of Hex Codes for colors
- Naming 99 colors,
hexcolor.html
is identical to medium except for missing the color `flesh'.
Hexcolor may be vestigal at UNI. It doesn't seem to be referenced from
other pages there.
- Color Name Dictionary: Netscape Color Names
-
Although there are many pages offering Netscape color
dictionaries, I could not find the definitive site.
ns3names.txt
is a 127 color subset of X11's `rgb.txt' with different values for
`gray' and `green'; and the addition of
`lightslate' and `mediumviolet' -- apparently
transcription errors of `light slate blue' and
`medium violet red'.
- Color Name Dictionary: HTML-4 Color Names
-
The HTML-4.0
specification defines 16 color names assigned to the full and half
coordinate RGB primaries:
Color names and sRGB values
 | Black = "#000000"
|  | Green = "#008000"
|
 | Silver = "#C0C0C0"
|  | Lime = "#00FF00"
|
 | Gray = "#808080"
|  | Olive = "#808000"
|
 | White = "#FFFFFF"
|  | Yellow = "#FFFF00"
|
 | Maroon = "#800000"
|  | Navy = "#000080"
|
 | Red = "#FF0000"
|  | Blue = "#0000FF"
|
 | Purple = "#800080"
|  | Teal = "#008080"
|
 | Fuchsia = "#FF00FF"
|  | Aqua = "#00FFFF"
|
Obviously, names were assigned to the colors. Red, lime, and blue
(and probably others) will lie outside of the print gamut.
- Color Name Dictionary: Hollasch colors.txt
- Steve Hollasch compiled 190 color names
into colors.txt.
More than 20 of its "whites" match the
Sinclair Paints colors in X11.
The clump of whites is visible in CIELAB
space and sRGB space.
The sRGB extremes and a couple of greens will exceed print gamuts; but
otherwise a reasonable effort.
`colors.txt' looks fine for light sources; and appears better
than `rgb.txt' for surface colors.
- Color Name Dictionary: Resene Paint Colours
-
Resene Paints Limited,
New Zealand's largest privately-owned and operated paint manufacturing
company, has generously made their "Resene
RGB Values List" available.
With over 1300 colours, Resene fills a large
volume of the CIELAB space uniformly. The
primaries (red, green, blue, cyan, magenta, yellow, white, black) are
absent, as they should be for physically realizable paints.
With so many names, most are specific to Resene. But their
organization and breadth of coverage will make them less confusing and
more useful than the other systems examined here.
Resene-rgb.txt is an X11 style color
dictionary which Augmenting rgb.txt
explains how to install.
The "Resene RGB Values List" is an excellent source for surface colors.
Resene colours in L*a*b* space
|
|---|
|
|---|
- Color Name Dictionary: Resene-2007 Paint Colours
-
While the Resene paint colours circa 2001 are an excellent source,
there are some technical flaws:
-
The whitepoint (the reference white) was set too high, causing many
colors to have RGB components of 255. [A color with an RGB component
of 0 or 255 is indicated in the catalog by a black X.] There are 113
duplications, many in the lightest colors. No paint is completely
reflective; so the whitepoint for a paint color should be less than
sRGB:255/255/255.
-
The blackpoint (the reference black) was set too low, causing many
colors to have RGB components of 0. No paint is completely
absorptive; but neither is any video display! So the low blackpoint
is a less serious problem.
-
Resene's paint colors are expressed in sRGB coordinates. Some cyan,
yellow, and magenta paint colors will be out of the sRGB gamut and not
reproduce faithfully from this source.
Since 2001, Resene
Paints Limited has acquired several other companies and
incorporated their colors into its
palette, which now
numbers over 2400 colors. Examination of
download_pencils.xls
reveals that both the whitepoint and blackpoint for their RGB values
have been moderated. There are still a few R, G, or B values of 255
and 0; but no more than one appears in a particular color. Some
singleton 0 and 255 values could arise as the result of trying to
assign sRGB values to colors outside of the sRGB gamut.
Nearly doubling the size of the "Resene RGB Values List" is not
desirable. So I extracted the new RGB values for only the existing
list in creating Resene-2007-rgb.txt.
Only three colors were lost in the process: "Cedar Wood Finish",
"Nutmeg Wood Finish", and "Teak Wood Finish". The number of duplicate
colors was reduced from 113 to 36.
The Resene-2007 catalog compares this
new Resene catalog with the 2001 version. Most noticeable is that
many reds and some blues are less saturated.
Resene-2007-rgb.txt is an X11 style color
dictionary which Augmenting rgb.txt
explains how to install.
The "Resene-2007 RGB Values List" is an excellent source for surface colors.
- Color Name Dictionary: NBS/ISCC Dictionary of Color Names
-
Kenneth L. Kelly and Deanne B. Judd.
"Color: Universal Language and Dictionary of Names",
National Bureau of Standards,
Spec. Publ. 440, Dec. 1976, 189 pages.
A compendium of colors and their names from academic, government,
textile, dye, paint, horticultural, and manufacturing sources would
appear to be the ideal source for surface color names.
In The NBS/ISCC Color
System, David Mundie has tried to computerize this information. At
the core of his system, The 267 Color
Centroids lists 267 colors, their Munsell renotations, and
hexadecimal sRGB values.
267 colors is more manageable than
Resene's 1300. It is enough surface colors for most computer
applications. And the centroid names are systematic, allowing users
to figure the name of a bluer shade without searching the catalog.
But there are problems with the centroids:
- Colors `115' and `129' are both named `Vivid
Yellowish Green'.
- `135 Light Yellowish Green 0.7g 7.4 5.2 #007BA7' looks
blue.
- `263 White 2.5pb 9.5 0.2 #FFC9D7' is very pink.
- `9 Pinkish White 5.8r 9.0 0.8 #F9DBC8' looks whiter than
`263 White 2.5pb 9.5 0.2 #FFC9D7'
- Using a program downloaded from http://munsell.com, spot checks of
the RGB values versus renotations do not agree with Mundie's data.
This may be the fault of the program -- its conversions are not
reversible.
I was able to locate a library copy of the NBS book. Quantitative
measurements are not presented -- all colors are specified using Munsell
renotations from Albert Henry Munsell's 1905 publication "A
Color Notation".
The fractional renotations in `Cent.html' are not from the book.
Each centroid is shown as an outlined area on Munsell hue charts. The
color dictionary section does not specify color distinctions finer than
the 267. It specifies color ranges with multiple
centroid colors, just as Mundie's dictionary pages do.
Starting through the `A's:
- the listing of `Acacia' as 150 is missing.
- `Agathia Green 114 131' should instead be
`Agathia Green 117 131'
- `Ageratum 227 should instead be
`Ageratum 253.
These human sorts of errors indicate that the dictionary was not scanned
and read (successfully) by an OCR program. The typography of the
dictionary section is sloppy. And use of the letters `l' and
`O' in close proximity to numbers would make OCR just about
impossible.
I checked the first 50 entries in the dictionary and found 3 errors;
there are over 5400 color names in total. Correcting all of it is a
daunting task, but it has been done (see next entry).
The centroids look good in in CIELAB space.
NBS Centroids in L*a*b* space
|
- Color Name Dictionary: NBS/ISCC Centroids
-
Kenneth L. Kelly and Deanne B. Judd.
"Color: Universal Language and Dictionary of Names",
National Bureau of Standards,
Spec. Publ. 440, Dec. 1976, 189 pages.
John Foster of the Texas Precancel Club
has greatly improved the
NBS centroids:
- Transcription errors from NBS Spec. Publ. 440 have been fixed.
- The duplicated colors have been separated.
- The Munsell coordinates have been reconverted using the
Munsell conversion program
from
Gretag-Macbeth.
White is now white and the pink bias is eliminated.
The NBS/ISCC catalog compares Foster's
NBS/ISCC with Mundie's, showing better balance and larger distinctions
between centroids.
Munsell has changed their standard several times in the last century,
so recreating the exact centroids of the 1950s edition of the NBS/ISCC
color dictionary would be difficult, if not impossible.
The nbs-iscc RGB-cube shows that the
named colors cover less than half of RGB color space. With the RGB
primaries being mapped to physically realizable surface colors, this
system does not have names for primary illuminants. Adding modifiers
such as "luminous" or "bright" would be a way to extend the system to
luminance.
NBS-ISCC-rgb.txt is an X11 style color
dictionary which Augmenting rgb.txt
explains how to install.
NBS-ISCC Centroids is a good source for
surface colors.
- Color Name Dictionary: NBS/ISCC Dictionaries of Color Names
-
Kenneth L. Kelly and Deanne B. Judd.
"Color: Universal Language and Dictionary of Names",
National Bureau of Standards,
Spec. Publ. 440, Dec. 1976, 189 pages.
John Foster has also digitized all 5411 color names in the NBS/ISCC Dictionary of Color
Names. This dictionary contains color names used by 32 commercial
and government organizations circa 1950. Many names map to more than
one color. Some are close in hue; others are not.
The 5411 color names map to just 267 centroids, but not uniquely.
Many names map to each centroid; but one name can also map to many
centroids. Capturing only the uncontested color names would reward
the obscure ones and avoid the common names we are most interested in.
Better would be to discard the outliers and combine the other colors
of each cohort. Where a color was midway between adjoining centroids,
the average would be closer to the original color than any centroid.
But the state of the art color metrics work only for close colors;
partitioning will be an algorithmic challenge.
Another approach would be to narrow the set to a specific source,
which is documented for each color name. There are paint, dye, and
ink makers among the sources, so each of those narrowed lists would be
similar in scope to the Resene paint colors.
Each list would have resolution limited to 267 colors unless the
multiple colors were close and could be averaged. Over 50 years old,
the list would not include colors achievable with new pigments or
processes. Each NBS/ISCC sub-dictionary listed here is marked with
its source letter-code (in parentheses):
- Color Name Dictionary: (M) Dictionary of Color
-
Maerz and Paul,
Dictionary of Color, 1st ed.
This idiosyncratic dictionary's 3064 names map to 4401 colors, using
223 of the 267 centroids.
- Color Name Dictionary: (P) Plochere Color System
-
Plochere Color System
This idiosyncratic dictionary for interior decorating has 1246 names
mapping to 1606 colors, using 206 of the 267 centroids.
- Color Name Dictionary: (R) Color Standards and Color Nomenclature
-
Ridgway,
Color Standards and Color Nomenclature
While using some modifiers, this primarily idiosyncratic dictionary
was intended for biology and botany. It maps 1096 names to 1602
colors, using 214 of the 267 centroids.
- Color Name Dictionary: (T) Descriptive Color Names Dictionary
-
Taylor, Knoche & Granville,
Descriptive Color Names Dictionary
While using some modifiers, this primarily idiosyncratic dictionary
intended for the "mass market" maps 673 names to 1612 colors, using
233 out of 267 centroids. Although 87% coverage of centroids is
good, each name mapping to an average of 2.4 colors reduces its
usefulness.
In technical terms, NBS/ISCC derived color dictionaries will not be
superior to Resene's set. And it is not clear that having more than
one color dictionary of 1000+ paint colors brings any benefit. But
the smaller color sets coded by other organizations could be of
practical importance:
- Color Name Dictionary: (TC) Standard Color Card of America
-
Textile Color Card Association [TCCA],
Standard Color Card of America and U.S. Army Color Card
The 237 names of this idiosyncratic dictionary map to 267 colors,
but only 129 out of 267 possible centroids are called out.
- Color Name Dictionary: (A) Dye Colors
-
American Association of Textile Chemists and
Colorists and Society of Dyers and Colourists
This combinatorial dictionary's 88 names map to 350 colors; 263 out
of 267 centroids are covered.
- Color Name Dictionary: (B) Colour Terminology in Biology
-
Colour Terminology in Biology,
H.A. Dade
This combinatorial dictionary seems to be derived from Latin
(eg. viridi-caeruleus). Its 128 names map to 789 colors, using 214
of the 267 centroids. But averaging over 6 colors per name means it
has little descriptive power.
- Color Name Dictionary: (F) Colors; (for) Ready-Mixed Paints
-
Federal Specification TT-C-595,
Colors; (for) Ready-Mixed Paints
This dictionary is a hybrid of combinatorial and idiosyncratic
terms. 134 names map to 174 colors, using only 73 out of 267
centroids.
- Color Name Dictionary: (H) Horticultural Colour Charts
-
Horticultural Colour Charts,
R. F. Wilson
Most names are of plants or flowers. 199 names map to 261 colors,
using only 95 of 267 centroids.
- Color Name Dictionary: (S) Postage-Stamp Color Names
-
Postage-Stamp Color Names,
William H. Beck
This dictionary is a hybrid of combinatorial and idiosyncratic
styles. 178 names map to 905 colors, using 223 of 267 centroids.
Despite covering 84% of the centroids, averaging over 5 colors per
name makes this dictionary uselessly unspecific.
- Color Name Dictionary: (RC) Rock-Color Chart
-
National Research Council,
Rock-Color Chart
This combinatorial field-work dictionary's 95 names map to 120
colors, using 89 of 267 centoids. This is less redundant than the
others. Unfortunately, the focus on rock colors will not make for a
general purpose dictionary.
- Color Name Dictionary: (SC) Soil Color Charts
-
U.S. Department of Agriculture
Soil Color Charts
This combinatorial field-work dictionary's 50 names mape to 176
colors, using 76 of 267 centroids. Again, its focus on soil colors
limits its general usefulness.
We desire a dictionary that is both specific (mapping each name to
fewer than 2 colors, averaged) and covers the NBS/ISCC centroid
colors. Only two had enough specificity to be of interest: the
rock-color chart (RC),which covered only 1/3 of
the centroids; and Colors; (for) Ready-Mixed
Paints (F), which covered even less. It appears that the
centroids themselves remain the most valuable part of the NBS/ISCC
effort.
| Catalog | sRGB Space | L*a*b* Space | Duplicate Colors | Source
|
|---|
|
|
|
| 287 of 627
| CNS syntax
|
- Color Name System: CNS color-naming system
-
Color bibliographies sometimes refer to A New Color-Naming
System for Graphics Languages by T. Berk, L. Brownston, and
A. Kaufman, IEEE Computer Graphics and Applications,
Vol. 2, No. 3 (May, 1982), pages 37-44.
The authors propose and critique several color name systems with the
goal of being able to name colors without charts or chips. Of the
ISCC-NBS centroids they conclude:
The main problem with this system from a user's point of view is that
there is no simple rule to determine which hue terms can be paired,
can modify or be modified, and in which order they must appear. In
order to use this system, one needs either the list of 267 distinct
names, the 31 color name charts, or the set of ISCC-NBS centroid
colors. The load on the casual user's memory is too great for this
system to be easily used.
One would expect they would create a system with fewer colors. But
their color-naming system has more than
double the number of color names (627), many of which are not
"realizable":
The syntax of CNS is orthogonal with respect to the three terms of a
color specification. All possible combinations are syntactically
correct, although some syntactically correct combinations may not be
realizable. For example, if a color is very light or very dark, it
cannot be fully saturated. Only colors of intermediate lightness can
be vivid. The maximum possible saturation of a color decreases as it
becomes light or dark. This is the reason for the common description
of the color solid as a sphere or double cone, rather than as a
cylinder.
As the Resene colours in L*a*b* space show,
the hull of surface colors is neither spherical nor conical; it is
grossly asymmetrical around the color-neutral (vertical) axis.
Their human factor experiments used color samples from
The Munsell Book of Color and the
Macbeth Colorchecker Color Rendition Chart.
Although they must have assigned colors to names in order to conduct
the experiments, the paper gives no such assignments, Munsell or
otherwise. They write:
In an application requiring computed color values, the computation can
still be carried out in the RGB, HSV, or any other appropriate system,
once the initial values have been determined.
Assignment of actual colors to the 627 names is apparently left as an
exercise for the reader.
- Color Name Dictionary: Color Naming Experiment
-
In 2003, Nathan Moroney presented
Unconstrained web-based color naming experiment,
at the SPIE/IS&T Electronic Imaging Symposium.
It describes a website he created
(http://www.hpl.hp.com/personal/Nathan_Moroney/color-name-hpl.html)
which invites visitors to submit names for 7 randomly chosen colors.
These are collected, vetted, and averaged to create a color dictionary
which can be queried at
http://h20325.www2.hp.com/blogs/color/archive/2007/10/29/4914.html.
Each query returns a color with its sRGB coordinates, a 3-level
indication of how popular the name was, and four nearby named colors
as well. By tracing the nearby colors, John Foster collected
871 named colors from the website.
Moroney's paper reveals that the randomly chosen colors have
coordinates selected from only 6 values which are uniformly spaced
from 0 to 255. But sRGB is not a perceptually uniform color space!
This results in the dark colors being severely underrepresented among
the colors presented for naming.
The CNE RGB-cube shows artifacts of the
limited dark samples: the darkest octant is nearly empty; and clumping
is visible along the half-intensity axes.
Compared with the Resene-2007
RGB-cube, the CNE RGB-cube is seen
to be distended from around the neutral (black-to-white) axis into
saturated primaries which lie outside of the print (reflective) gamut.
The paucity of dark colors also hides some inherent technical issues.
Coordinates for dark color names are likely to show more relative
variation than the light colors because displays have more variation
in contrast ratio, blackpoint, and dark color balance than they do in
the highlights.
CNE-core-2007.txt is a subset of the
CNE dictionary containing only the 212 "widely used", and "less
commonly used" colors.
Looking at its RGB-cube shows the
darkest octant emptiness is worse than for the full CNE, perhaps
reflecting that agreement between subjects is harder to obtain for
dark colors.
The Color-Naming-Experiment could be improved by rewriting the webpage
to:
- generate random colors with component resolutions of tens or
hundreds of values, not just 6 values;
- generate random colors in a more perceptually uniform color space
than sRGB, such as CIEL*a*b* or CIEL*u*v*;
- set the background color of the page. Moreover, using a black
background would harness the human visual system to compensate for
some of the differences in blackpoint, improving the agreement of
darker colors.
All that being said, the
Color-Naming-Experiment dictionary
is still an improvement over X11. It is good for light sources and
mediocre for surface colors.
- Color Name Dictionary: saturate
-
I created the saturate
colors from assignments shown on the rim of a chart showing
"Approximate Colors on CIE Chromaticity Diagram" credited to the
book Computer Graphics by Francis S. Hill (Macmillan,
1990, pg. 572). Colors from red to green to blue are monochromatic.
The colors from blue to purple to red have chromaticities interpolated
between 400.nm blue and 700.nm red.
Each of these colors is out of the RGB709 (and hence sRGB) gamut. If
you think gamut-mapping is easy, try mapping these. Use
`saturate' colors for light sources only.
Using technical means, the genealogical relations between online color
dictionaries have been elucidated. Most online color dictionaries
(larger than 16) are derived from X11's "rgb.txt" without proper
attribution and without an explicit license or copyright.
Most online color dictionaries suffer from poor technical quality,
assign names to RGB colors outside the sRGB gamut and the gamut for
surfaces, and severely under-represent the darkest octant of the RGB
color cube.
The Resene,
NBS-ISCC Centroids, and
saturate dictionaries are developed which:
- have explicit copyright and license;
- have no name conflicts between them;
- are specific to surface colors or lighting;
- have colors distributed to equalize perceptual distances between
them; and
- are derived from matching physical samples (except saturate).
Augmenting rgb.txt gives instructions for
installing Resene-2007 and NBS-ISCC Centroids as X11-style color
dictionaries.
Copyright © 2001, 2002, 2003, 2004, 2005, 2006, 2007 Aubrey Jaffer
I am a guest and not a member of the MIT Computer Science and Artificial Intelligence Laboratory.
My actions and comments do not reflect in any way on MIT.
|
| | Color
|
| agj @ alum.mit.edu
| Go Figure!
|
