QUOTE (djellison @ Dec 12 2004, 09:37 PM)
...perhaps a combination of layers multiplied by something etc etc.
I've been working on that as well, trying to expand out my color work from the 6 filter into the 3 filter...but with limited success. As it stands right now, I've got numbers based on conversion from an L2 L5 L7 spectrum (linearly interpolated between L2 -> L5, L5-> L7) into CIE XYZ and CIE XYZ to RGB.
These numbers are based on no whitepoint conversion
Red = 1.394 * L5 - 0.505 L7 + 0.314 L2
Green = 1.026 * L5 - 0.079 * L7 + 0.005 * L2
Blue = 0.037 * L5 + 0.932 * L7 - 0.010 * L2
I've done some work to figure out the proper whitepoint by using the sundial images, though that falls apart quickly due to dust accumulation (I'd LOVE to know what whitepoint the JPL images are using)
Here's those same figures based on a Bradfield whitepoint conversion to D65 based on the whitepoint of X = 97, Y = 100, Z = 75 which is my best calculation/guess as to the whitepoint of Spirit.
Red = 0.938 * L5 - 0.530 * L7 + 0.250 * L2
Green = 0.972 * L5 - 0.117 * L7 + 0.009 * L2
Blue = 0.123 * L5 + 1.21 * L7 + 0.003 * L2
And the figures based on a Bradfield whitepoint conversion to D65 based on the whitepoint of X = 97, Y = 100, Z = 85 which is my best calculation/guess as to the whitepoint of Opportunity.
Red = 1.043 * L5 - 0.515 * L7 + 0.266 * L2
Green = 1.002 * L5 - 0.104 * L7 + 0.009 * L2
Blue = 0.067 * L5 + 1.108 * L7 - 0.001 * L2
now for the "and here's why they're wrong"
As you said, L2 is a HORRIBLE approximation of red, and barely even contributes to the red channel even with the interpolation. Here's why:
I found/made these up for a recent discussion over on another message board, but they illustrate the point well :
As you can see, L2 isn't even on there (its center is well beyond the red center of human vision, with light coming through the L2 only contributing about 1% of the red that we perceive), although L3 4 and 5 and 6 all fall under the 'red' curve, which is why L5 contributes so much to the red part of the equations above.
Also, linear interpolation is a very bad way to interpret the spectrum in between the filters we have for most of the 3 frame color. As is polynomial...and splines. Basically, any straight mathematical interpolation does a very bad job of matching up with the real data since L2 is such a bad approximation. My current project is to try to get a statistical approximation using the 6 frame color to fit a curve to each individual filter, so at least it'll be a best guess as to the values in between 2 filters based on real ground truth. I'll let you know if that works.
And lastly, as I said...in terms of matching to the JPL color, it would be fundamentally easier if we could know what whitepoint is being used for the conversion. They might very well (and probably should) be using a different whitepoint at different times throughout the mission, since whitepoint is based on the illuminant, and as your graphs of the Tau values show, the brightness of the sun/sky varied greatly even over the first 90 sols. My best guess at this point is that the whitepoint at the beginning of the mission was much more 'red' than later on, since the sun brightness increased relative to the dust-laden sky brightness over those first 90 sols. This would mean that the surface illuminant was less dominated by the red sky and more dominated by the direct sunlight as the mission progressed.
I hope you are able to get use out of these. If you wanted to use a whitepoint other than the ones I picked, let me know, and I'll calculate the channel math based on it.