Doug Kerr
Well-known member
In discussions of color spaces and the like we encounter the terms chromaticity, chrominance, and chroma. Sometimes it seems as if these all describe the same thing, the author just getting progressively more tired of spelling it all out. In fact, they all refer to different, although closely related, things.
Two aspects of color: one outlook, luminance and chromaticity
One way to look at color that relates well to overall human perception is to consider color to have two aspects, luminance and chromaticity. We can think of luminance as inditing the brightness of the light (there being a slight wrinkle there, which we need no fret over this morning). We can think of chromaticity as the aspect that distinguishes red from green (the sub-attribute of hue) and red from pink (the sub-attribute of saturation). This is not a "recipe" outlook on a color space.
There are several ways that the chromaticity can actuality be specified, always in terms of two values (giving us three values altogether to specify the color of interest).
Two aspects of color: another outlook, luminance and chrominance
An alternate technical way to look at color is in terms of the two aspects luminance and chrominance, a "recipe" form. We think of making up light of the color of interest by starting with a ration of "white" light (and of course we understand that "white" must have some arbitrarily-defined meaning) and adding to it a "colorant dose", which makes the result some non-white color.
The luminance aspect of the color description tells us the amount of the white base light in this "recipe", and the chrominance aspect tells us the nature, and amount, of the colorant dose.
There are various ways to specify the chrominance, always in terms of two values.
The reader may note a paradox here. After we add the "colorant" light, won't the overall luminance be higher, no longer consistent with the amount of white light, which we said was dictated by the luminance? No - the colorant isn't "light" - it is just colorant. The colorant is "impotent", luminance-wise. Hard to visualize? Sure. This is only a mathematical concept, not really a way to "make" light, and the math is set up to make it work that way.Here's the big difference between chromaticity and luminance. Suppose we have an instance of light with a certain luminance and a certain chromaticity. It will also have a certain chrominance.
We attenuate it, as with a neutral density filter (perhaps by "one stop"). Then:
• The luminance is half what it was
• The chromaticity is unchanged (it has, for example, the same hue and saturation)
• The chrominance is half of what it was
We can perhaps understand the third of these with a little analogy. We imagine that we are having mixed some paint of a custom color. In this analogy, the quantity of paint will represent the luminance of an instance of light. (I know that doesn't seem quite right, but that's what is needed to make the story work!)
I we want a gallon of paint, we start with a gallon of "base white" and add to it, according to a recipe, perhaps one ounce of "red oxide" and two ounces of "cadmium yellow", together, the "colorant dose".
But if we instead wanted one quart of paint, we would start with one quart of "white base" and add 1/4 ounce of red oxide and 1/2 ounce of cadmium yellow. To get the same "chromaticity", we need to add a smaller amount of colorant to a smaller amount of base.
Now, for chroma
First, note that the perceived luminance of a color can be reliably calculated from a linear equation involving the values r, g, and b, which are the relative amounts of the three primaries that are (actually or conceptually) combined to make the color of interest. (These are on a "linear" basis, for the benefit of those trying to get ahead of the story.)
In the sRGB color space (and many others of the RGB family), the values R, G, and B are not those three values. Rather, they are typically those values raised to a power (perhaps 0.45). The original purpose of this, in the color TV ancestors of these color spaces, was to outguess the nonlinear response of the three color "guns" in the display CRT. Later, it was recognized that there are other advantages of that arrangement (although they would have been better attained with a slightly different arrangement).
But in the NTSC color TV system, these nonlinear R, G, and B values weren't transmitted. Rather, two different "aspects" were derived and represented by separate electrical signals. One was a linear combination of R, G, and B, which is something like luminance but not really - its is not even a non-linear representation of luminance. (I'll spare you the mathematical illustration of this.) It came to be called "luma". That name (rather than "luminance") implied two things:
• We were talking about an electrical signal, not the number it represented
• Oh, yes, it has a nonlinear premise, and in fact isn't even exactly a nonlinear form of luminance
The other was derived (in a slightly complicated way) from the differences between R and Y and between G and Y. Together, those two form a two-dimensional quantity that was something like chrominance, but not quite. (It was not even a non-linear form of chrominance.) It came to be called "chroma". That implied two things:
• We were talking about an electrical signal, not the number it represented
• Oh, yes, it has a nonlinear premise, and in fact isn't even exactly a nonlinear form of chrominance
Now, if we have a TV signal representing a particular color, and we "fade that toward black", what happens:
• Luma decreases
• Chroma decreases [we'll talk a little later about exactly what that means]
Today, we have a prominent digital color space, the sYCC color space, that operates on a luma-chroma basis. It is derived from the (nonlinear) values R, G, and B. Its coordinate Y (luma) is a linear combination of R, G, and B - not luminance, not even a nonlinear form of luminance, but "something like it" - and we can conveniently call it "luma". (It of course is not an electrical signal, though - just an 8-bit number.)
Its coordinates Cb and Cr are calculated as B-Y and R-Y, respectively. They together form something that is not chrominance, nor even a nonlinear form of chrominance, but something like it. We can conveniently call it "chroma". (It of course is not an electrical signal, though - just two 8-bit numbers.)
Another usage of "chroma"
We run into the term "chroma" in another place. I have mentioned that "chroma" is two-dimensional (in the mathematical sense - two values are needed to specify it). There are two ways to specify it.
a. As made up of two coordinates "at more-or-less right angles" (on a chromaticity diagram).
b. As an amount ("the "magnitude" of the chroma) at a certain angle (on a chromaticity diagram).
If we say, in some scenario, "the chroma is decreased", that means:
A. Under scheme a, both coordinates decrease, proportionally.
B. Under scheme b, the magnitude decreases
For any given luma, a "decrease in chroma" corresponds to a decrease in saturation.
It is for that reason that the term "chroma" is used in connection with some color spaces to mean something like saturation (the important qualifier underlined just above usually being an implied part of the definition, often not realized).
Two aspects of color: one outlook, luminance and chromaticity
One way to look at color that relates well to overall human perception is to consider color to have two aspects, luminance and chromaticity. We can think of luminance as inditing the brightness of the light (there being a slight wrinkle there, which we need no fret over this morning). We can think of chromaticity as the aspect that distinguishes red from green (the sub-attribute of hue) and red from pink (the sub-attribute of saturation). This is not a "recipe" outlook on a color space.
There are several ways that the chromaticity can actuality be specified, always in terms of two values (giving us three values altogether to specify the color of interest).
Two aspects of color: another outlook, luminance and chrominance
An alternate technical way to look at color is in terms of the two aspects luminance and chrominance, a "recipe" form. We think of making up light of the color of interest by starting with a ration of "white" light (and of course we understand that "white" must have some arbitrarily-defined meaning) and adding to it a "colorant dose", which makes the result some non-white color.
The luminance aspect of the color description tells us the amount of the white base light in this "recipe", and the chrominance aspect tells us the nature, and amount, of the colorant dose.
There are various ways to specify the chrominance, always in terms of two values.
The reader may note a paradox here. After we add the "colorant" light, won't the overall luminance be higher, no longer consistent with the amount of white light, which we said was dictated by the luminance? No - the colorant isn't "light" - it is just colorant. The colorant is "impotent", luminance-wise. Hard to visualize? Sure. This is only a mathematical concept, not really a way to "make" light, and the math is set up to make it work that way.
We attenuate it, as with a neutral density filter (perhaps by "one stop"). Then:
• The luminance is half what it was
• The chromaticity is unchanged (it has, for example, the same hue and saturation)
• The chrominance is half of what it was
We can perhaps understand the third of these with a little analogy. We imagine that we are having mixed some paint of a custom color. In this analogy, the quantity of paint will represent the luminance of an instance of light. (I know that doesn't seem quite right, but that's what is needed to make the story work!)
I we want a gallon of paint, we start with a gallon of "base white" and add to it, according to a recipe, perhaps one ounce of "red oxide" and two ounces of "cadmium yellow", together, the "colorant dose".
But if we instead wanted one quart of paint, we would start with one quart of "white base" and add 1/4 ounce of red oxide and 1/2 ounce of cadmium yellow. To get the same "chromaticity", we need to add a smaller amount of colorant to a smaller amount of base.
Now, for chroma
First, note that the perceived luminance of a color can be reliably calculated from a linear equation involving the values r, g, and b, which are the relative amounts of the three primaries that are (actually or conceptually) combined to make the color of interest. (These are on a "linear" basis, for the benefit of those trying to get ahead of the story.)
In the sRGB color space (and many others of the RGB family), the values R, G, and B are not those three values. Rather, they are typically those values raised to a power (perhaps 0.45). The original purpose of this, in the color TV ancestors of these color spaces, was to outguess the nonlinear response of the three color "guns" in the display CRT. Later, it was recognized that there are other advantages of that arrangement (although they would have been better attained with a slightly different arrangement).
But in the NTSC color TV system, these nonlinear R, G, and B values weren't transmitted. Rather, two different "aspects" were derived and represented by separate electrical signals. One was a linear combination of R, G, and B, which is something like luminance but not really - its is not even a non-linear representation of luminance. (I'll spare you the mathematical illustration of this.) It came to be called "luma". That name (rather than "luminance") implied two things:
• We were talking about an electrical signal, not the number it represented
• Oh, yes, it has a nonlinear premise, and in fact isn't even exactly a nonlinear form of luminance
The other was derived (in a slightly complicated way) from the differences between R and Y and between G and Y. Together, those two form a two-dimensional quantity that was something like chrominance, but not quite. (It was not even a non-linear form of chrominance.) It came to be called "chroma". That implied two things:
• We were talking about an electrical signal, not the number it represented
• Oh, yes, it has a nonlinear premise, and in fact isn't even exactly a nonlinear form of chrominance
Now, if we have a TV signal representing a particular color, and we "fade that toward black", what happens:
• Luma decreases
• Chroma decreases [we'll talk a little later about exactly what that means]
Today, we have a prominent digital color space, the sYCC color space, that operates on a luma-chroma basis. It is derived from the (nonlinear) values R, G, and B. Its coordinate Y (luma) is a linear combination of R, G, and B - not luminance, not even a nonlinear form of luminance, but "something like it" - and we can conveniently call it "luma". (It of course is not an electrical signal, though - just an 8-bit number.)
Its coordinates Cb and Cr are calculated as B-Y and R-Y, respectively. They together form something that is not chrominance, nor even a nonlinear form of chrominance, but something like it. We can conveniently call it "chroma". (It of course is not an electrical signal, though - just two 8-bit numbers.)
Another usage of "chroma"
We run into the term "chroma" in another place. I have mentioned that "chroma" is two-dimensional (in the mathematical sense - two values are needed to specify it). There are two ways to specify it.
a. As made up of two coordinates "at more-or-less right angles" (on a chromaticity diagram).
b. As an amount ("the "magnitude" of the chroma) at a certain angle (on a chromaticity diagram).
If we say, in some scenario, "the chroma is decreased", that means:
A. Under scheme a, both coordinates decrease, proportionally.
B. Under scheme b, the magnitude decreases
For any given luma, a "decrease in chroma" corresponds to a decrease in saturation.
It is for that reason that the term "chroma" is used in connection with some color spaces to mean something like saturation (the important qualifier underlined just above usually being an implied part of the definition, often not realized).
