[Radiance-general] Re: measuring color
Greg Ward
gward at lmi.net
Wed May 12 17:57:56 CEST 2004
This is an interesting thread. I just wanted to add a couple of notes.
First, you should check out the products available from X-rite for
spectral measurement:
http://www.xrite.com/product_overview.aspx?Line=24
I have used their spectrophotometers in the past, to good effect. I
think they're prices are competitive, but not what I'd call "cheap."
Second, a quick explanation of what macbethcal does. Macbethcal takes
a captured image (scanned, photographed, etc.) of a Macbeth
ColorChecker chart and attempts to derive a transformation that takes
this capture to the correct values known for the Macbeth chart. It
first computes an overall response function for each channel using the
6 gray patches at the bottom of the chart. Then, it computes a 3x3
color matrix based on a least squares minimization of error on the
unsaturated colors. Saturated colors are allowed to go "out of gamut"
as needed to find a reasonable solution.
In my experience, this process works well for flatbed scanners and some
capture methods, but sometimes does poorly with certain digital
cameras. I do not know why. As Alexa and Jack pointed out, macbethcal
does not account for changes in the illumination source -- this is
assumed to be part of the capture process (i.e., a scanner with a
builtin source), and since the purpose is to derive a transform that
recovers colors under this souce, this is appropriate.
If you wish to simulate general reflectance spectra under different
illuminants, I suggest you go with a full spectral measurement and
follow the procedure outlined in the following paper:
http://www.anyhere.com/gward/papers/egwr02/
Third, I wanted to point Alexa and other interested readers to the
following reference, which explains the parts of human vision pcond
attempts to simulate and the algorithms behind it:
http://radsite.lbl.gov/radiance/papers/lbnl39882/tonemap.pdf
I'll be the first to admit that it is not a perfect simulation of human
vision, and this is an ongoing area of research. (Try a websearch on
"color appearance models" to get an idea.) However, I think others
will agree that pcond can be useful, and is an improvement over the
standard "linear with clamping" method for image display.
-Greg
More information about the Radiance-general
mailing list