I'm using argyll to profile a couple of scanners, driven by Vuescan.
1) Epson V500 flatbed
2) Nikon LS-5000ED
The target is the Lasersoft ISO12641-2 with its three parts.
I'm using a shaper matrix profile approach as I found that using a CLUT can
wildly accentuate noise from grain in lighter colors, esp skin tones. But
this is an aside.
When working with the target, I've found scanin to be incredibly touchy
about getting a mapping. A crop difference for a fraction of a percent can
make a difference, and scanin will reject what look like perfect fit from
the diagnostic output TIFF. I've seen this problem discussed in the forums
from 10 years ago with no clear answer, except that when users futz around
with adjusting the crop it eventually can me made to work. I haven't tried
to debug it, but it's a major PITA. There is some edge case which throws
scanin and it's not grokkable based on factors under the user's obvious
control.
As to performance...
The Nikon is clearly superior in resolution to the Epson, with the Nikon
producing resolution close to 4000 spi as claimed (based on Lasersoft Air
Force 1951 target) and the Epson barely eking out 1800 effective spi, which
requires some HF attenuation to make it look good (PS Smart Sharpen works
well). Epson bald faced lies about it's scanners capabilities (6400 spi
optical is a patently ridiculous claim)
Surprisingly, while the Nikon's far-superior resolution is obvious, it
makes a less difference in photos that I expected. it's color performance
that matters a lot and Nikon does not have a clear advantage in color or
tonality.
Which brings me to my question:
After I got scanin to accept the targets, I noticed was very large peak
errors in profile generation with the Nikon — like 25 dE or sometimes
higher. The Epson peak error is about 9. At first I thought this might be
caused by large amounts of bloom in the shadows with the Nikon, due to very
dirty optics. This old Nikon mosdef needs to be cleaned. But the worst
error patches are not darks next to brights, as I had suspected. The worst
errors are in blues in the second target. Especially at patch G38, in the
blues that surround this patch, and the cyan group above. (The histogram
from profcheck shows a pronounced hump with a very distinct separate tail.)
For patch G38, I went through an exercise of comparing the profcheck output
with Photoshop and it all added up except that visually there are wide
swatches of patches with a far more noticable difference than these dark
blue patches.
I overlaid the profiled Nikon with the profiled Epson and discovered much
less visual difference in their rendering of patch G38 that many other
patches. Further investigation revealed that this patch is a film blue
which is out of gamut for sRGB and also for my Dell WCG (Adobe + P3)
displays.
It seems that the Nikon and Epson see very saturated blue component very
differently and this affects colors all over the map, especially yellow :).
My explanation to myself so far is that Epson sees a blue that the Nikon
designers might have punted on because the know that most displays will not
show it. I haven't yet compared a map for an Epson 4800 printer but I have
printed a corrected IT8 side-by-side and while the colors are distinctly
different, neither makes the other look wrong. They just have different
personalities.
Am I on the right track with my thinking? Is it known, for example, that
the filters Nikon applied made such a compromise on blue response, and that
a large peak error in blues comes with the territory of some scanners and
cameras?
Any thoughts appreciated
/wire
As I compared the results from the Epson and the
