.
SBFAQ Part 2: Color Discrimination
Marc Green
2.1 How many colors can the eye distinguish?
2.2 Then how come my documentation says that I can create 16.7 million colors?
2.3 What factors affect color discrimination?
2.4 How many different color categories do people distinguish?
2.5 How many hues does the eye distinguish?
2.6 How many gray levels can the eye distinguish?
2.7 How can I get more than 256 gray levels out of my computer by using color?
2.8 How many saturation levels can the eye distinguish?
2.1 How many colors can the eye distinguish?
There are several ways to answer this question. The most sensitive method is to place colors side-by-side and ask whether they appear the same or different. However, there are too many possible color combinations to determine the actual number of distinguishable colors by direct experiment. Several sources (Halsey and Chapanis, 1951; Kaiser and Boynton, 1989) have used the limited existing data combined with some theory to estimate that under best conditions, the eye can distinguish one million colors - combinations of hue, saturation and brightness. Other estimates, however, put the number as high as 7 million (Chapanis, 1954).
In less optimized conditions, the number of discriminable colors is much smaller. If the color samples are separated in space or in time, the number of distinguishable colors falls precipitously. The number also falls with smaller size, shorter viewing time, lower illuminance, and several other factors.
In electronic screens, CRT, plasma, LCD, etc., the technology further limits the number of distinguishable colors. Electronic displays can't produce a large percentage of the visually perceptible colors. Technical references sometimes say that there 50,000 discriminable CRT colors, but it is unclear where this number originated. The lowest estimate I've seen is 20,000.
2.2 Then how come my documentation says that I can create 16.7 million colors?
You can't. A 24 bit RGB system does not create 16.7 million colors; it simply creates 16.7 million (256 x 256 x 256) different machine states. Remember, color is manufactured in the head, not the machine. Even assuming that it is creating 16.7 million different spectral compositions (sets of wavelengths), it does not follow that the computer produces 16.7 million distinguishably different colors. The number is likely in the tens of thousands.
- The eye does not distinguish 16.7 million different colors, anyway.
- A CRT monitor can't produce all of the million colors that the eye can distinguish. Specifically, computer and TV screens can't produce the most saturated colors. Moreover, ambient light falling on the screen further lowers saturation and shrinks the number of colors which the display can produce.
- Even within the range of color that it can produce, 24 bits still quantizes the color space too coarsely in some regions. Colors may jump by more than one JND (just noticeable difference step) because discriminability is not uniform throughout the RGB color space of monitors.
In sum, many of the 16.7 million states produce indistinguishable colors while there are many of the million distinguishable colors which "true color" can't produce.
Note added 1 Aug 08: I have seen this section cited to mean that increasing color resolution to 36 or 48 bits would produce no image quality improvement. This is incorrect. True, the range of colors producible is a function of the display device and does not change. However, increasing the number of bits per color improves the color resolution (8 bit=256 steps, 12 bit=4096 steps) so that successive colors do not jump more than one JND. The end result is smoother gradations and more realistic looking images, especially at low brightness where the tendency to quantize too coarsely most occurs.
2.3 What factors affect color discrimination?
The number of distinguishable colors in real world situations is difficult to specify. Many
environmental factors can improve or impair color discrimination:
- Spatial separation: Color discrimination is much poorer when the samples are separated so that they don't form a border. The greater the separation, the worse the discrimination.
- Number of dimensions. Colors will be more discriminable if they differ in hue, saturation and brightness than if they differ in only 1 or 2 dimensions.
- Spectral location. Hue discrimination varies across the spectrum. Normal viewers are most sensitive to hue changes around yellow and at the border between blue and green. Color discrimination is poorest for colors from the edges of the spectrum, red and violets. Some color deficients, however, have different color discrimination abilities.
- Size: Discrimination is poorer for small objects. Hue, saturation and lightness/brightness discrimination all decrease. The effect is greatest for yellow and blue. (Watch out when you compare carpet and paint samples! Colors which look similar as 3 inch squares in the catalogue may appear very different when covering the floor and walls at home.)
- Saturation: Hue discrimination becomes worse as colors become less saturated.
- Brightness: Hue discrimination declines at lower brightness.
- Time separation: Color discrimination is poor for colors viewed successively and compared in memory. If more than a few seconds elapse, people can easily discriminate only 10-12 colors. Discrimination is best when the colors are central members of each of the 11 basic color categories.
- Learned Color-Object Familiarity: People remember colors better when shown familiar objects (a red apple, green leaf, etc.) rather than simple light patches. Conversely, color is more poorly remembered if a familiar object has the "wrong" color - a blue apple or a green heart.
- Retinal location: Color discrimination is best when objects are imaged in the fovea - you are looking directly at it - and falls as the image is seen further in the periphery. Discrimination first degrades for red and green and then blue and yellow before failing completely. This effect is largely irrelevant for CRT monitor viewing since it only starts at a location so peripheral that it is beyond area a person uses when viewing a computer screen at normal distances. If projecting an image on to a large screen, the peripheral insensitivity could be a problem. A cautious designer could minimize loss of peripheral color discrimination by using highly saturated colors, yellow and blue and large objects.
- Brief presentation: Short durations seriously impair discrimination of similar colors. The effect is greater with reds and greens and smaller with blues and yellows.
2.4 How many different color categories do people distinguish?
For many designers, the number of psychophysically discriminable colors is less important than the number of color categories which people readily perceive. Viewers will more quickly and efficiently group objects in the same color category and distinguish objects in different categories.
Since many distinguishable colors fall into the same category (there are many shades of green, etc.), the number is going to be relatively small. Berlin and Kay performed a famous linguistic study which concluded there are 11 basic color distinctions that fall into three classes:
- achromatic color terms: black, gray, white
- primary color terms: red, green, blue, yellow
- secondary color terms: brown, orange, purple, pink
However, there is a hierarchy within this set of colors. Virtually all languages have words for black and white. This could be interpreted to mean that they are the most basic colors in some sense. If there is a third color term, it is invariably red. As languages add even more color terms, there is a predicable order: green, then blue and yellow, then brown and finally the secondaries and gray. It is unclear whether this hierarchy represents a natural ordering of perceptual significance, but it is worth noting that some graphic designers believe that you should use only three colors: black, white and red.
Within each category, moreover, there are good and bad examples. The good examples fall in the center of the category while the bad ones fall close to the border of two categories. Unique hues are the best category examples for the perceptual primaries. People typically remember the good examples better and name them faster than poor examples. This suggests that visual design should use central category members to accentuate important distinctions, especially if colors must be remembered.
Further, people obviously make many more subtle distinctions in color than merely the 11 cited by Berlin and Kay. The National Bureau of Standards once published a list 7,500 different English color names, but they are not fundamental categories. In some cases, however, color names have a special meaning. For example, there is a set of standardized color names for web designers who don't like talking in hexadecimal.
Later work by Kay and McDaniel (1978) suggests that some of these other terms may represent highly salient, if not basic, color names. They studied Polish, Swedish, Spanish (Spain) and English (US) speakers and found that people of these languages commonly used a term for beige, sometimes more often than some "basic" color names. Most speakers also frequently used terms for colors such as navy, turquoise, sky blue, gold, silver and violet. Kay and McDaniel also developed a theory suggesting that all color terms were merely combinations (intersections in fuzzy set theory) of the 6 focal/landmark/opponent process colors - red/green, blue/yellow, black/white.
2.5 How many hues does the eye distinguish?
The number of just noticeable difference steps (JND's) in the spectrum is about 150 (Wright, 1947). There are some additional extraspectral colors (purple) and colors which can only be made by mixing/contrast (brown), so the total number of hues is somewhat greater.
2.6 How many gray levels can the eye distinguish?
Several studies have converged on the same answer, roughly 450. Eight bit color (256) will
produce banding artifacts if mapped over a wide brightness range. The situation is complicated
by the inequality of step sizes across brightness levels. At the low end steps should be very
small while the high end needs larger steps (Weber's Law). It is unlikely that 8 bits really
supply 256 gray levels, let alone 450. In general, the steps at the low end are too large, so
each gray level jumps more than one JND while at the high end, the steps are too small, so
successive gray levels appear identical.
But there is good news. CRT monitors have a defect, nonlinear gamma, that partly corrects this
error by automatically decreasing step size at the low end and increasing the number of discriminable
gray levels. Unfortunately, there is also bad news. Both the size of human brightness JND's and the
screen gamma change with viewing conditions, so the number of steps that is really available also
changes and is somewhat unpredictable.
2.7 How can I get more than 256 gray levels out of my computer by using color?
Even a 24 bit system only supplies a maximum 256 levels of gray since it uses 3 eight bit numbers to control intensity. However, there is a simple trick which can improve gray level resolution.
The relative intensity of the colors green-red-blue form a rough ratio of 4-2-1. The exact ratio will vary from monitor to monitor, but the order is always the same - green is most intense, then red, the blue. Now suppose that you have a gray at (red=50,green=50,blue=50). Normally, the next step would be (51,51,51). But you can create 6 intermediate steps between these two levels:
(51,51,51)
(51,51,52) = +1
(51,52,51) = +2
(51,52,52) = +3
(52,51,51) = +4
(52,51,51) = +5
(52,52,51) = +6
(52,52,52)
Most people worry that these intermediate steps will cause a color shift. In fact, the steps are so small that they produce little or no change in perceptible hue.
2.8 How many saturation levels can the eye distinguish?
Starting at white and going out to a pure spectral color, there are gradations in apparent saturation. However, some hues appear inherently more saturated than others. The most saturated yellow still appears pale compared to saturated red, saturated green and especially saturated blue. As a result, the number of distinguishable saturation levels is smaller in yellow than in the rest of the spectrum. One study concluded that there are only 10 saturation steps around yellow with the number gradually rising as wavelength increased or decreased. The lowest wavelengths, blue-violet had about 60 steps while the red reached about 50. Viewers will find small differences in blue, violet and red saturation highly discriminable while small differences in yellow saturation will be hard to detect. Green and orange are in the middle
Back to Top
A division of 2057949 Ontario, Inc.
Copyright © 2024 Marc Green, Ph. D.
Home Page
Contact Us
