non Green really complements non Red (sort of)

By Roy Zuniga

In previous posts I confirmed my initial thesis that the traditional colors artists consider as complements (Red/Green, Orange/Blue and Yellow/Violet) are not actually complements in the pure color space of the computer. I showed how none of the colors that are 180 degrees apart in the HSB color space align with the artist’s complements. And we saw how the computer complements actually do neutralize each other. So in a purist sense, is the computer right and artistic color is wrong?

Wait a minute, aren’t artists by definition color experts? How could they be so off? I postulated that this may be due to a number of factors, including color shift (the illusion that colors shift hue as they are tinted), and the colors available to artists (which are not as saturated or bright as what you can show on a screen). For example, the chart below shows how fully saturated, bright colors (center stripe) tend to shift in hue as they are tinted white (top) or black (bottom).

You’ll notice that yellow looks decidedly greenish as it is tinted toward black, even though the hue number doesn’t change. The computer tells us that both the saturated color in the middle and the one close to black in the yellow band are in fact the same number (60). What changed is the brightness; saturation was maintained constant at 100%.

Other colors play similar tricks on our eyes: orange looks redder as it gets darker, magenta shifts into violet, and violet tugs at blue in the darker tones. Cyan lists toward green and green warms up considerably in the darks. In the lights, orange and green both look yellow; and a few of the lights in the red to violet range seem identical to the untrained eye; same for the blue greens. What appear as 12 distinct colors in the saturated range become about 5 or 6 distinguishable colors as they approach white or black.

How does all this account for the gap between tradition and the algorithmic complements? I decided to do a little experiment, which is very subjective by nature. What follows is not a proof; it’s a plausible explanation at best.

First, I selected typical artist colors for main colors, red, orange, yellow, green, blue and violet (Viridian, Cad Red, Cad Yellow, Cobalt Blue, etc.) Since various manufacturers produce paints that look different even though they have the same name, this is an inherently subjective exercise. I squirted out a few and got an optical range to work with. I adjusted the ‘pure’ colors in my color calibrated monitor to more or less match the blotches on the white canvass paper. In other words, I created my equivalent of an artist’s red, orange, yellow, green, blue and violet on the computer, and used the same blending feature to blend these adjusted colors along the traditional pairings.

The result was quite interesting. When I examined the hue number for my adjustments, it turned out some had shifted. For example, pure yellow is #60, while my eyeball artist’s yellow is #56; orange went from #30 to #32 (no big deal there); Red didn’t shift hue at all and neither did blue; green leaped by nine from #120 to #129; and violet went up 8 ticks from #270 to #278. The Red/Green duo grabbed my attention next, since there seemed to be convergence to more closely align with the computer complements. I decided to push the envelope to see if the would converge further.

Using the same blend tool and applied to my ‘artist’ colors, blue was mixed with green and violet with red to get some intermediate colors. Sure enough, I found I could accept colors that were closer to my target 180 degrees of opposition. I picked two new colors, as shown here.

Re-running the blend, the result was a further narrowing of the gap, showing how from the artist perspective, red and green can be loosely considered complements. My optical green lost 60% of it’s brightness from the pure green. As we saw above, that would have shifted its appearance towards red. So to compensate, the hue was moved towards blue. Red, as it loses brightness definitely pulls towards orange. To compensate I instinctively adjusted toward violet. Consequently, the two converged. The blended colors you see below are not true red and green according to the computer. Ask any artist, however, and she will likely say they are red and green – and perception is everything!

Since these mislabeled colors are very close to the true computer complements, it is plausible that empirically colorists did see them neutralizing each other in the past (you now know better, of course). Thus the ‘artist’ adjusted colors are only 23 degrees off the 180 degree opposition required to be true complements. This is much less than the 60 degrees departure we see in the pure saturated version of green and red. In other words, in the pure computer model, red (#360) is two complement bands away from green (#120). In the optically adjusted model, we can get away with saying they are less than one band away – and in the fuzzy world of mixing pigments, the distinction may be academic. Thus we account for the origin of the myth that red and green are complements! The colors known as red and green behave closely to how complements behave, i.e. neutralizing each other. 

Some lessons we can take away include:

1. A plausible explanation of how the classical complements model was conceived as colorists mixed pigments they had at hand.
2. An understanding that if we want to neutralize colors with their true complements, we have a new chart to look at and pick from.

copyright 2009 roy zuniga

Computer Complements - Confirmed

by Roy Zuniga

On my last post titled 'Compliments of your Computer', I noted that the traditional artists' color complements model does not align with color space in our computers. Using the HSL color space broken into 12 colors evenly divided in the numerical values yielded complements that were not the traditional Red-Green, Blue-Orange, Yellow-Violet. I noted the need for a scientific proof that the new proposed 'normalized' 12 step model did in fact result in neutralization of colors. This afternoon, before picking up my brushes, I decided to use a vector drawing program, Adobe Illustrator, and specifically it's ability to blend two colors in intermediate steps, to prove or disprove the thesis. The result confirmed the thesis, as follows.

Adobe Illustrator doesn't have the HSL model which uses numbers up to 255 for hues. Instead it has HSB which uses a 360 model. So I had to divide the 360 into the 11 intermediate steps to get the same chart -- but the results were the same. To go to the next step and find if the middle blend is in fact neurtral, I then used the blend tool with 7 intermediate steps to oppose complements. For example hue 0 was opposed to hue 180, and this was done for all 12 colors, with Saturation and Brightness dialed up to 100% for all. The resulting middle blend color was in all cases 50% brightness and 0% saturation -- in other words a neutral grey.

If I shifted any colors and opposed them with something other than its 180 degree complement, some non-neutral color was obtained in the middle. Thus the theory is confirmed. Red is neutralized by a turquoise color, not green, for example.

In this HSB 12 color breakdown, however, yellow and orange didn't show up as nodes. So using the RBG controls I dialed in the traditional primaries Yellow, Red, Blue and converted to HSB to chose their complements. The result was the following five opposing sets of colors. As expected, the center colors are also neutral grey. These stripes would fit in the above wheel to the right or left of blends shown. This chart shows more clearly what the complements to your favorite primaries are, according to the computer color space.

Please note that the original file was exported for a web color space for posting on this blog, and the colors slightly shifted due to the limited color pallet of 'web colors'. However, I did verify the colors on the original file, and the middle colors are in fact 50% brightness at 0% saturation (making the hue number irrelevant, i.e. all hues are neutral grey at those percentages).

So the thesis stands. How this translated to artists color practice remains TBD. I would encourage artists to experiment. Keep in mind that artists colors are not always at the highest brightness and saturation for any given hue. A useful next step is to map commercial artist colors to color nodes based on this model, and then do manual blends to confirm or repudiate the theory with artist colors. We could start with numbering the chart above and finding commercial equivalents. Of course the challenge will be to compare a physical pigment with a computer model because monitors are calibrated differently, so perceived color will vary. In the mean time, it won't hurt to use this chart as your color complements chart. If you want a PDF version, just send me an email (zuniga.roy@gmail.com), and I'll get you a chart.

One final note: I'm not an expert in this area, but we need a way to objectively 'read' a commercial artists' color and correlating that to a value in the above color space.

Happy Painting!

copyright 2010 roy zuniga

Compliments of Your Computer

by Roy Zuniga

I was on my computer composing a little matrix of colors to aid the distribution on a palette colors in what I call an opposing complements layout, starting with coolest opposing hottest at the top:

I like this palette because it puts colors that neutralize each other in direct opposition. As I found the colors in the computer color picker for my reference card, I realized that the increments between color numbers were sometimes regular (about 11) -- especially in the range of Yellow/Violet and Green/Red -- and sometimes much larger, as in the Turquoise related hues (which lack complements in the historical model). The computer has a color model (Hue, Saturation, Lightness or HSL) built in (one of several). So I took the highest number (255) and divided by 12 = 21.25. This yielded both the traditional colors we're used to, as well as a few others that fall into main steps (Red is both 0 and 255 because color space wraps):

Shuffling these regular step colors into my old opposing complements model, I noticed there were anomalies, i.e. half steps as well as gaps (shown in black):

The traditional model is shifted and lacks some complements (like those for Turqoise-Blue (149) and Green-Turquoise (106)), and has falsely paired some colors with half-step colors (for example, Orange-Yellow (32) and Red-Orange (11)). So I normalized the colors according to full steps and opposed them so the difference in numbers was always 128:

We'll call this the 'Normalized 12-Step Opposing Complements' model that arranges complements according to the computer color space. This yields a very pleasing matrix that also makes sense mathematically.

Could it be that the traditional theory of complimtary colors that survives in our art books is an evolution of observation limited by the pigments available and is not accurate.

This may explain why mixing direct complements in the old model sometimes shifted tinted colors into adjacent color spaces and didn't strictly neutralize. The traditional model is close enough that subtle shifts in low intensity colors were not perceived or ignored. Moreover, since artists pigments rarely have colors that directly align with the steps, we're used to mixing several colors. In practice, the artist's eye overrides the theory, which is always just an aide to getting started. For these reasons, the traditional model has gone unchallenged in artist's circles. I would encourage artists to validate or invalidate this 12-Step Normalized Complements model.

Now, I'm going to stop messing with the computer and start creating art -- Happy Painting!

copyright 2009 roy zuniga