That’s your first mistake. You have to season to taste the settings to get a match.
"Season the settings to get a match?" It's been an amusing thread...
In Europe, SpectraView Profiler includes a display monitor with hood, custom calibration software developed by basICColor, and each display is individually certified. (
http://www.necdisplay.com/supportcenter/monitors/spectraview2/faq/) The European market is arguably more demanding than the North American market.
LUT profiles are widely accepted as giving more accurate profiles, both visually and statistically, than matrix profiles, because theoretically, a LUT profile is better able to handle non-linearities in the system - that's why 3-D LUT profiles were developed.
From the same NEC reference above:
QUESTION: Why doesn't SpectraView offer an option to make full 3D LUT type ICC/ColorSync profiles in addition to the current shaper/matrix profiles?
ANSWER: The LCD panels used in the SpectraView displays have excellent color linearity characteristics and can be characterized using the much simpler shaper/matrix profiles. Accurate 3D LUT profiles require a minimum of around one hundred measurements (and up to several hundred) to generate the necessary data and offer very little advantage despite the large increase in measurement time. Additionally there are some compatibility issues with various software applications when using 3D LUT type display profiles.
So NEC says LUT profiles "...offer very little advantage." Yup. And sometimes that "very little advantage" is important to the discriminating user. In Europe, it's a no-brainer - use BasICColor to make a 3-D LUT profile. Compatibility issues with LUT monitor profiles? Name one! Not Photoshop, LR, or any of the software that I've used for years.
As for the differences Mark found between using a matrix profile and LUT profile, first off, he's making comparisons using the same hardware - both monitor, computer, and colorimeter, so those variables have been eliminated. The NEC software, as Mark pointed out, is limited by the fact that SpectraView (North America) is unable to make a LUT profile, so it's handicapped to begin with. No option, no choice. As was pointed out by Czornyj, he could make a matrix profile and LUT profile with BasICColor and compare both to prove the point, but undoubtedly you'd then state that the BasICColor matrix profile is not as good as the NEC matrix profile, or find some other problem with the comparison. The real issue is which software package is capable of giving more accurate results.
SpectraView and BasICColor each have their own calibration routine that uses DDC to access the internal monitor LUTs, but it's not possible to compare the accuracy of each calibration, because there is not access to the calibration data for both. Presumably, Mark was calibrating to a gamma 2.2 TRC with SpectraView, and to a CIECAM02 TRC with BasICColor. Again, SpectraView II is just not up to the task of using a more modern, visually accurate TRC (although the European BasICColor software is), as using a CIECAM02 TRC may give a better visual match to his prints. After all, that's why it was developed.
Next, in profiling, a 16-bit LUT using CAT02 chromatic adaptation (from CIECAM02) is more likely to give an accurate profile, as determined visually or statistically, than a simple matrix-based profile using ____ (none?, von Kries?, linear Bradford?) chromatic adaptation. (AFAIK, SpectraView II doesn't even tell you what, if any, chromatic adaptation it uses. Do you know?) It is widely accepted that matrix based profiles are smaller in filesize, but are somewhat less accurate, whereas LUT based profiles are larger in filesize and generally more accurate (but may sacrifice smoothness). In Mark's case, determining which gives a more accurate profile can be verified visually (which he did!), as well as by comparing the statistics on both profiles using the same hardware. But I can already hear you: "The colorimeter is not a $50,0000 reference spectrophotometer!!!"
Even if the hardware measurement device has some inherent inaccuracy or "offset," it will not affect this statistical comparison. There is no reason to expect that a profile that gives
larger errors will somehow be "more accurate" than one that gives small errors, given that the same device is used to make the measurements that were used to build the profile as well as to verify the measurements; a perfect profile would show an error of zero for all data points or colors, and a poor profile will show large errors. Of course, there may be some
systematic error due to the inaccuracy of the colorimeter, causing the colors to be "off" compared to a reference instrument, but there is NO WAY that a profile with a LARGER verification error, built from data from the same measurement device, is going to have greater absolute accuracy than a profile built with data from the same device that has a LOWER verification error. The profiling app has
no way to know whether or not the data it receives from the colorimeter is accurate, or to compensate for an "inaccurate" colorimeter. The profiling apps's job is to make a profile that can output color with minimal error - as measured by the device that generated the data used to build the profile. That error can be measured - preferably with deltaE2000 - for comparison purposes.
Verification routines are useful to compare different profiling applications (or settings within one app) to see which gives the lowest error, with the understanding that the absolute accuracy
may be off. Of course, if the accuracy of the device is off a lot, it's visibly apparent. If two or more measurement devices give comparable results, (say, an Eye-One Pro and a Display2) and images shown using the profiles look appropriate, then the absolute error of the devices is likely to be small. In addition, if the measurement device is traceable to NIST, as with a certified Eye-One Pro, any inaccuracy of the device is likely inconsequential in real-world use. This is, in fact, the basis for
certified monitor calibration by SWOP and GRACoL and the basis for virtual proofing. (
http://files.idealliance.org/certifications/cert_docs_2010/idealliance_proofing_cert_19h.pdf) Of course, DeltaE 2000 should be used for this verification and certification, because the older delta formulas are simply not accurate enough and do not correlate well with visual errors. The reason for this is discussed at the end of the Idealliance reference.
NEC claims that the shaper-matrix profiles from SpectraView (North America) are "good enough." Mark's experience was that
for him, they were not. Mark has an excellent eye for color, and he's very careful in testing and he is not one to jump to more complicated or expensive solutions without good reason to do so. The matrix profiles that he made with SpectraView visually failed, and the LUT-based profiles made with BasICColor give better results, visually and statistically. I think there is a reason that the European market demanded certified panels and BasICColor. If the NEC matrix profiles are good enough for you (or anyone else), fantastic! That doesn't mean that they're good enough for everyone.