Stephan Bourgeois wrote:
I then used a video camera + tungsten light source and, to my surprise, got a better result, but still with a color cast. (Well, not much of a surprise, tungsten wasn't filtered to daylight, I used the white balance on the camera.)
A color cast is typical when the input system really doesn't respond spectrally like a human/CIE observer.
I then used a digital camera (Kodak DX6440) under direct sunlight and got an *excellent* result, with almost no color cast. This camera doesn't have a "raw" mode (only sRGB JPEG), and there is some clipping on the primary colors, as one would expect in the sRGB space. No doubt I will do this again with a camera that gives more control and more range... I have only tested the profile with "photographic" images, I haven't tested it against very saturated color patches that may reveal errors due to the clipping.
My interpretation for those results is :
a. the light source could be a more important factor in color rendition of these devices than the spectral response of the RGB filters. The scanners I used offer the worst light source i.e. cold cathode tube, that looked very "spikey" in a handheld spectroscope.
The spectral sensitivity of the input system is the product of the light source and the sensor filters, so both are equally important. Of course with a camera you can easily change the light source, giving you a mechanism for moving the input system in the direction of being colorimetric.
b. the scanner seems to clip highlights when using a paper with fluorescent whitening agents. The clipping is different with a camera.
Hmm. It means that the scanner is clipping a "reflectance" over 1.0. It is often the case that scanners clip quite early, since many users expect that their paper white input to translate to perfect white on output - they don't want to see any background. For use as a substitute for a colorimeter, it's good to be able to turn this processing off, and get wider range readings.
Here are my questions: 1. Please comment on this.
2. What would be the difference between using a colorimeter (eg. Color mouse CM2C) and a digital camera with a good quality light source (xenon strobes)?
Very great. A colorimeter could be expected to have close to CIE responses, will be even across the patches, and has a very high dynamic range.
3. Can I use the peak error and average error reported by the "profile" command as an indicator of the accuracy of the color measuring device?
No, you would have to compare the readings you get against a known higher accuracy device, such as an instrument. The errors reported by "profile" simply indicate the degree of fit between the computed profile and the original measurement points. A bad fit indicates inconsistency amongst the measurement values.