The Color Dimension

(Go back to main Profiles page)

March 26, 2006

The "zone system" is a way to know and fully utilize the tonal properties of black & white film and printing papers for optimal rendering of images in black & white. It was conceived to provide photographers control over image tonality and it has clear analogs in color photography for control over tonality. But when working in color, the color dimensions themselves beg for similarly effective solutions. I have a unique solution for fundamental control over the principal one amongst those dimensions: that of the amount of color in an RGB mode image. I have used it for nearly a decade while creating my own images and I believe it works better than any other method available. This essay discusses the topic and why I think my method, using my chroma variant profile sets for RGB working spaces, is the best way.

Digital color images can be thought of as large number of discreet colors, positioned in a 3D space like CIE L*a*b*. This device-independent space has white positioned at the top and black at the bottom, with a straight line of grays connecting them. Around this gray backbone are then found all the non-gray colors, including every hue (e.g. red, orange, yellow, green, cyan, blue, magenta) arranged in a longitudinal fashion around the grays. The closer to the height of white a color is, the lighter it is, and the closer to the height of black the darker. The further from the gray axis a color is, the more colorful it looks.

Traditional, chemical systems for making color photographs have generally been severely lacking in control over anything beyond color balance and overall exposure level of the image. Selecting subject matter for its contrast, selecting film for the tonality inherent in its characteristic curve, alternating between two emulsion types of a given color paper, making masks for original color film for controlling contrast, and dodging and burning were essentially the only ways to exert control over the tonality of a color print. Even when practiced to the greatest extent, these controls are unable to provide anything approaching full control over tonality.

Each subject demands its own tonal solution when a photographic interpretation is made of it, and nothing short of the control offered by digital imaging systems, with their RGB curves and L* curve tools among others, is sufficient to achieve optimal tonality from the vast range of subject matter which the world presents to photographers. The human visual system affords the equivalent of instant, highly effective control over not only this dimension of what we see, but over the white balance, the zoom (viewing angle), and even the sense of saturation of the color. Photography doesn't do these things automatically, for the most part, so photographers must do them themselves if they wish to make pictures which actually look as much like the subject as is possible, given the physical limitations of the print medium or the display.

But what about the other major dimension of the color spaces? Curves give us great control over the vertical position of the color points of an image -- affording exquisitely fine control over tonality when coupled with a carefully crafted workflow. But images also require interpretation in the horizontal dimension. I would submit that the extent of colorfulness of an image is nearly as unlikely to be automatically optimal as is the extent of contrast in any given part of an image's tone curve, although the tonal problems do predominate, as a photographer faces the problem of rendering a given scene optimally.

When RGB curves are used to control image tonality, the effect on the image is much the same as when the same curve is applied to a subject by color film having that curve shape. As the slope of the curve is increased between any two points on the horizontal axis (any small region of the tone scale of the image) the image contrast (light/dark contrast) for that region increases, and the image's color (color contrast one might call it) for that region increases in exact proportion. For this reason, if one is working with a scan of a color transparency, which has imposed it's own very strong curve shape onto the image being captured, one can use RGB curves to undo what the film did to a large extent, and return the image to a state much closer to the actual luminous character of the subject, as desired, but within the limits imposed by the film's latitude onto the dynamic range of the subject.

When an L* curve tool is used on an image in Lab mode, the curve slope does not affect the image color at all, and so tonality can be altered independently of color. RGB curves tools, when used in Photoshop's Luminosity mode, have the ability to alter tonality almost independently of color as well. Another kind of interesting tool is the Chroma curve tool, which allows excellent control over the colorfulness dimension of the Lab color space, but this tool is found in Linocolor and NewColor software for Heidelberg scanners (and perhaps in ColorQuartet software for ScanView scanners as well, if not others), but not in Photoshop, as of CS2. The a* and b* curves in Photoshop's Lab mode can be used to mimic a Chroma curve tool if used in a certain, rather awkward way.

The Lab color model is just one mathematical model that seeks to isolate the dimension of colorfulness from the dimension of tonality. Technically the sensation of colorfulness is called exactly that -- colorfulness -- but we often refer to it as saturation in photography. Technically, within the realm of digital imaging, saturation refers to the particular effect yielded by various RGB color models (e.g. HSL and HSV) as the variable of saturation (S) is changed, as by using the saturation slider in the Hue/Sat tool with RGB mode images in Photoshop.

The object, for us in adjusting the amount of color in an image, is to adjust it without adjusting apparent tonality. Since no color model yet devised, let alone in use in imaging software, actually shifts image color in exactly this way, the best option we have still involves some apparent shifting of tonality, but much less than the other options do.

The Chroma dimension of Lab (the distance from the gray axis in a*/b* units) is this best model for altering apparent colorfulness among any available in our workflows today. Adjusting the saturation dimension of either the HSL model used in Photoshop or the HSV model used in Live Picture, for example, causes image tonality to shift several times more than adjusting Chroma does. Another way to state that is to say that the L* dimension fairly closely matches our sensation of lightness or luminosity throughout the range of visible colors, therefore when changing the Chroma (a* and b*) dimensions of Lab data, which doesn't change the L* dimension, apparent luminosity is preserved very well. In Chroma the reds shift in apparent lightness the most.

In practice, the application of control over the amount of color in an image, independent of the tonality of the image, involves a good deal more than just the character of the color model used. Choosing an optimal method involves considering all the factors of the systems available.

The Hue/Sat adjustment layer and tool in Photoshop allow us to adjust Saturation in the HSL color model when working with RGB images. Making a fairly large adjustment of Saturation this way, in either a positive or negative direction, readily reveals the problems inherent in the HSL model when the image is a fairly colorful one to begin with, especially when the image contains several hues. Not only do image tones shift lighter or darker when they would ideally not, but they shift enough to ruin the tonality of the image altogether in many cases, making the normal mode of the tool essentially useless as the basis for an overall system of control of the colorfulness of images.

The Hue/Sat adjustment layer and tool in Photoshop can be used in saturation mode, causing it to behave in a more usable fashion, with lessened tonal shifting compared with the normal mode, however the tonal shifting is still considerably greater than with the adjustment of Chroma.

Additionally, when any RGB image has its saturation increased with the Hue/Sat layer/tool in any tool mode, some image colors are likely to become blown out because their numerical values are driven to zero or 255, causing a complete loss of image detail in one or more color channels in those colors, and an accompanying shift in hue. The image's red, green and/or blue values are moved toward the walls of the RGB cube of the working space as saturation is increased. I regard inadvertently blowing out colors as one of the capital sins of fine imaging (which can easily occur when converting into RGB working spaces and which one always needs to be wary of during those conversions and other image editing operations).

If an image has its saturation decreased with Hue/Sat, the region of the RGB space covered by the image colors shrinks, but the space does not, causing a steadily increasing quantization error compared to what would be optimal for that image -- although this is a trivial consideration — if the image has more actual bits per channel than eight or ten and those steps are evenly spaced throughout the tone scale.

Another way to change image colorfulness in Photoshop is to convert the image from RGB to Lab mode, then use the Hue/Sat tool, and then convert back to RGB. In this case, the "Saturation" slider of the Hue/Sat tool actually alters the Chroma, resulting in a much higher quality of control over colorfulness. But there are a number of problems with this approach:

1) Only the RGB image layer can be converted into Lab, not RGB adjustment layers, so the image layer has to be copied, then converted and adjusted, then inserted back into the layer stack, and you can't see the effect of any such layers as you're deciding how much to adjust the chroma.

2) The conversion of 24-bit data from any RGB working space to Lab and back causes significant quantization error. This damages both the critical light/dark dimension, reducing the number of steps in the image between black and white, and the number of steps between gray and maximum saturation. A 24-bit image in a smaller RGB working space such as ColorMatch or sRGB could lose more than 75% of its discreet image values when making one such trip from RGB to Lab and back. If dithering is enabled in the Color Settings dialog in Photoshop 6 or later, it is used for RGB to Lab and Lab to RGB conversions and masks the underlying quantization fairly effectively by introducing noise. Conversions of data in 16-bit form, even if they were originally 8-bit data, are well protected from quantization damage in the RGB to Lab to RGB round trip and dithering is therefore never used.

3) Adjustments made in this way can't easily be made many times, so it's even more impractical to make repeated adjustments as you experiment to find the right amount of color for an image as it's evolving.

4) When converting back into the RGB working space after having increased Chroma in Lab mode, the image colors are likely to clip, because they may no longer all fit within the original RGB space, just as when increasing Saturation with the RGB mode Hue/Sat adjustment layer/tool.


Fortunately, there is yet another way to adjust colorfulness, which I devised many years ago, to alter the chroma of images in RGB mode, by using what I call chroma variants of my RGB working spaces (Chrome Space 100, J. Holmes*, Ekta Space PS 5, J. Holmes, and my five new DCam profiles) or the popular Adobe 1998 and ProPhoto RGB spaces. The chroma variants are specially constructed ICC profiles with some unique attributes. Chroma is the dimension of Lab space which approximates colorfulness.

As you may know, ICC profiles work in a color managed workflow by establishing equivalences between all possible numerical values of RGB pixels and values of either the CIE L*a*b* or CIE XYZ color spaces. By having two such equivalences, the values of pixels in one space, say a scanner or camera space, can be changed so as to yield the same actual color in another space, say an RGB working space. And then again the values of a file in an RGB working space can be changed to new values so that the colors will remain the same when viewing on a monitor or when being printed (within the limitations of the printing system and the conversion process). Therefore, when a profile is assigned to an image (Image > Mode > Assign Profile... in Photoshop 6 through CS, and Edit > Assign Profile... in Photoshop CS2), the new profile defines the colors anew and they change, because Photoshop instantly re-builds and re-applies the source simulation that converts all open images from their source profile to the monitor profile. Or the soft proofing simulation, if it happens to be enabled at the time, which is also rebuilt in a flash.

If you assign different RGB working spaces to an image in Photoshop and watch how it changes on screen, you will see the tonality of the image shift (gets lighter or darker) when the two RGB working spaces have different gamma curves. And if you watch the colors in the image, they will shift as the nature of the RGB working space's gamut changes. If an image has sRGB assigned to it (a small RGB space designed to mimic the gamut of one kind of HDTV) and you then assign a giant space like Kodak's Pro Photo RGB, you will see a considerable increase in saturation, in addition to the lightening which results from the change from sRGB's approximately gamma 2.2 tone curve to Pro Photo's gamma 1.8 tone curve.

My chroma variants do not shift either the color balance of the image or the tonality of the neutrals of the image, nor do they shift the hues of the image, when assigned in place of the master working space profile for which they were built. They only change the chroma, and they do so quite uniformly and with minimal shifting of lightness. But they also change the size of the gamut in exact proportion to the change in chroma -- so it's impossible to get clipping by increasing saturation with the chroma variants. And when used to lower the chroma, the space shrinks so there is an appropriate minimization of quantization error too! That is, the number of discreet colors in the image file remains precisely the same, regardless of which chroma variant you assign.

What's more, the chroma variants can very quickly and easily be assigned to the image to change its colorfulness at any time in Photoshop, any number of times, with no damage to the image whatsoever. The image's colors are merely being redefined so that in conversions to your screen (to the monitor profile) or to prints (to the printer profile) the two equivalences involved simply line up differently.

And since the tonality of my Chrome Space 100 profile was carefully and deliberately designed to match the output of a well-linearized printer, the conversion damage going from any profile in the Chrome Space 100 set of profiles (master or chroma variant) into the profile for such a printer does approximately the least quantization damage possible, helping to preclude avoidable posterization in a print.

This aspect of building RGB working spaces (attempting to match the tone curve of output devices) has been almost universally overlooked by other designers of such spaces. Only sRGB uses a special tweak of the curve near black to better match printer behavior than any Gamma curve can, although in the case of sRGB, the tweak is too small for an ideal printing result, and it does as much harm as good because sRGB has become, as it was intended to be, the default space of images posted to the Web, and as such it would be best if sRGB had a tone curve which is most readily matched by computer displays calibrated to precisely gamma 2.2, not to the sRGB tone curve which is very much like gamma 2.2 until it gets very near black.

My chroma variant sets and their master profiles are for sale directly on my web site at http://www.josephholmes.com/profiles.html. Please see that page for more info on the variants that are available.

To use them, you convert from the source profile (scanner or digital camera) into one of the master profiles, then simply assign variants as desired to change the image's chroma.


Thank you!


— Joseph Holmes, Kensington, California

http://www.josephholmes.com



*until recently this space was called Ektachrome Space, J. Holmes, but when I overhauled it in 2005, I changed the name to keep it from being confused with my other space, which is often referred to as Ekta Space. I have always given Ekta Space away, but not the other profile, nor the chroma variants. It is partly for this reason that Ekta Space has become an industry standard working space. It is available for free download on my web site, along with a Read Me file about it, at http://www.josephholmes.com/profiles.html.