Ben Goren wrote:
So, based on the recent (and past) threads about camera profiles, I'm now seriously thinking about building my own ``ultimate'' target. What I've come up with so far:
* A white ``patch'' that's essentially a piece of paper backlit by a $15 slave flash that will establish the white point (should be R=G=B=255 in the native camera output).
I think you need to think that though a bit more. Since you are attempting to characterize the camera, why should 255,255,255 be some sort of anchor point ? What XYZ does it correspond to ? Looking at a flash is highly likely to add to glare, particularly from anything surrounding the chart or from elements within the lens. Some sensors might behave very badly (streaking etc.), and some may even be damaged.
* A selection of hand-applied artist's pigments (perhaps pastels?) that I'll select and apply with the help of a local art instructor; the selection will include colors spectrally similar to those of a ColorChecker plus any that the instructor considers difficult or especially important to reproduce. If I can find any pigments that are close to monochromatic primaries, I'll be sure to include those.
The problems with this are mainly: Uniformity - if it isn't uniform, you may get large discrepancies between instrument measurements and camera measurements. Stability - it not very useful if you have to keep creating reference files because the pigments change too rapidly with time. Statistical representation - implicit in creating a calibration, particularly a matrix, is that the test chart spectra represent similar statistics to the colors of images in the real world. If they don't, then the profile's fit compromise will be tilted in a way that isn't optimal for real world images. ie., to do this well requires some work. A possible cheat is to use paint swatch samples, since they use a variety of pigments and are reasonable uniform. There may still be problems with stability, and there will certainly be problems covering a reasonable gamut, since popular pain colors are typically rather muted.
* If I can't get a spectrally-flat gray out of the iPF8100, then either a pigment or cut-up gray card or something-or-other to use as a gray balance reference.
Something carefully selected is needed for spectrally flat neutrals. Pure K print on a good substrate may be OK, but it depends on the black ink involved. CMY etc. is not going to be flat enough.
* And a glossy black something-or-other (piece of plastic, or maybe printed black with a glossy clear coat applied) to give an indication if glare is a problem.
I'm not sure how that will help. It may simply add glare. The black trap may give a better indication.
First, does that sound useful / sane?
It's a lot of work to do right. Buying a ColorChecker SG is sounding faster and cheaper :-) Building a black trap, obtaining a good white reference, and establishing uniform illumination seem worthwhile. Graeme Gill.