Pixelation

General => General Discussion => Topic started by: Tourist on March 30, 2011, 06:06:51 pm

Title: Munsell data
Post by: Tourist on March 30, 2011, 06:06:51 pm
I downloaded a copy of the Munsell color charts (http://www.cis.rit.edu/mcsl/online/munsell.php (http://www.cis.rit.edu/mcsl/online/munsell.php)) and converted them to RGB.  Similar to here: http://casoilresource.lawr.ucdavis.edu/drupal/node/201 (http://casoilresource.lawr.ucdavis.edu/drupal/node/201) if you want to look at doing it yourself.  I didn't follow his exact steps, but the results should be similar.

The problem is that instead of a nice simple palette, the data set has over 4000 entries.  Filtering the data to those that fit in RGB space and even removing the very darkest colors (<10% luminance), the set still has about 1500 or so.  Assuming I did the math right.  It's still too large, all of the software I have only handles palettes of 256 colors at a time.

So, I'm not quite sure what to do next.  Any suggestions?  Anyone want to see the data plotted in some images?

Tourist

edit: Found a free pre-computed set here: http://www.wallkillcolor.com/ (http://www.wallkillcolor.com/).   Although they use illuminant C instead of d65.
Title: Re: Munsell data
Post by: Ryumaru on April 03, 2011, 06:26:54 am
I don't know how useful munsell is to digital artists. It's primary goal, above simplifying colors into hue, chroma, and value, is to give people a way to accurately communicate about color with those 3 values. In traditional art, that would allow one person to theoretically mix the same color every time, or for two artists to describe that color to each other through common terms. But we already have a system like that in digital, as every color we see is comprised of RGB values.
Title: Re: Munsell data
Post by: Tourist on April 04, 2011, 01:48:34 am
You've got a good point.  Since computers use RGB, storing and communicating data in RGB makes the most sense.  Also, the Munsell nomenclature is clunky.

After poking at the data for a week, I see some interesting patterns.  In the Munsell data, different hues behave differently.  Pure RGB data values (and systems derived from pure RGB, like HSV) treat all hues in the same fashion.

For a constant chroma and hue, Munsell data tends to go from a saturated dark to an unsaturated light value (using the HSV definition of saturation = (max-min)/max ). 

Now the interesting bit.  For the red ramp (5R), the difference between the largest component (red) and the smallest component (blue) is almost constant at all values (light to dark).  This holds for all chroma, although the constant is different for each chroma. That is, for a given chroma, max-min = constant, not (max-min)/max = constant.  This suggests that the strength of a red-dominated color depends primarily on the redness, not the saturation. 

This relationship does not hold for the green ramp (5G).  As the colors in the green ramp grow lighter, the value (max-min) increases, and it increases at a fairly linear rate.  I think this is due to green being the largest component of luminance.
In any case, the green color behaves in a different fashion than the red.  I find that interesting.

I'm not sure if this sort of thing can lead to describing the Munsell set with simple math.  If nothing else, the Munsell data shows the relationships between colors rather well.  The specific RGB numbers are less important than the general trends in the data, I think.  For the chroma, anyway.  I'm not sure about the choice of hues, and Lab does a superior job of calculating luminance.

Still poking at the data set.

Tourist

edit:I may have to double check my math.  The 5R ramp in csv file linked above does not have this property, although it does use a different illuminant than the d65 standard I'm using.  Hmm....

edit2: Well, that was a chunk of time.  My math was mostly good, needing a minor change to a matrix.  The results only shifted by one or two points of RGB.  I have to assume the csv file differences are due to the illuminant conversion.