Ad
Ad
Ad
Pages: [1]   Bottom of Page
Print
Author Topic: Testing spectrophotometers  (Read 3698 times)
Rhossydd
Sr. Member
****
Offline Offline

Posts: 1965


WWW
« on: April 05, 2011, 04:05:33 AM »
ReplyReply

Following on from a discussion in another thread about the absolute accuracy of spectrophotometers. I’m intrigued by the idea that it ought to be possible to assess a spectro’s useful photographic performance* and how it deteriorates in performance over time (yes, I know that sending it away for re-calibration would be the preferred option, but you’re unlikely to get any feedback with respect to if the recalibration was actually needed or how much the instrument had deteriorated in performance since manufacture or last calibration).

What could one use as an absolute colour reference ? preferably something affordable.
CC24s seems to be too variable in manufacture to consider a realistic choice.
The DCC sg is allegedly more accurate and consistent.
A measured IT8 might fit the bill ?
Any other ‘standard’ colour charts/patches that perform better ?

What might one use as a long term reference ?
Just to check change in performance over time materials that don’t fade are obviously crucial. However just to test change over time, there’s no inherent need for samples to be any specific colour.
Ceramic tiles ?

Anyone else tried this sort of thing ? or reported credible results ?

*Reality check;
Yes, I do understand this wouldn’t be ‘absolute’ in scientific terms, but might be useful for ordinary photography within the sort of tolerances most people expect.
Yes, there will be issues with needing to calibrate luminance measurement against a known standard.
Yes, there may be limitations in accuracy if you don’t have a wide enough gamut for test patches.
Yes, I know that the charts mentioned for using as a reference are themselves subject to manufacturing variations and may fade over time too.
Logged
Ernst Dinkla
Sr. Member
****
Offline Offline

Posts: 2867


« Reply #1 on: April 05, 2011, 06:44:30 AM »
ReplyReply

With the Spectrocam some years old I became worried about that issue. At that time (7 years ago ?) I made a tablet of several 3mm thick acrylic samples, different hues, some greys, a black one and a white one. Ask a company that sells acrylic sheets (Perspex, Plexiglass, Acrylite) for samples like that, 5cm square attached to a thin chain usually. Made measurements and kept the list stored with the tablet in a black envelope.. Today I would add a piece of PTFE to get a better white. I made new measurements when I got the Display 1 Basic last year. They are not certified colored ceramic tiles, Siberian opal glass or pressed PTFE powder tiles but degradation in the dark of cast acrylic is a very slow process.


met vriendelijke groeten, Ernst

New: Spectral plots of +250 inkjet papers:

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
Logged
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3752


« Reply #2 on: April 05, 2011, 10:05:42 AM »
ReplyReply

What could one use as an absolute colour reference ? preferably something affordable.

I would assume that any material that is capable of reflecting light in a relatively spectrally neutral way (in the relevant spectral band, and preferably with a Lambertian reflection pattern) would suffice. The manufacturer's calibration would presumably consist of electronic linearization/calibration, and loading a lookup table for a number of known reference materials (laboratory grade=expensive). Since we are dealing with a spectrophotometer, small reflective fluctuations over time in narrow spectral bands would be an indication of selective drift. For wavelength calibration, some material with known absorption peaks (e.g. Erbium Oxide) would be required.

Ceramics seem logical candidates because they can be relatively inert, pure/consistent in composition, and dense/thick enough to avoid background reflection. They are also relatively easy to (keep) clean if their surface structure has a smooth finish. The surface structure is however not very Lambertian in its reflection pattern, so light angle needs to be constant.

Something like the BabelColor White Target seems suitable for this purpose, and is available at (reasonably) acceptable cost.

Quote
CC24s seems to be too variable in manufacture to consider a realistic choice.
The DCC sg is allegedly more accurate and consistent.
A measured IT8 might fit the bill ?
Any other ‘standard’ colour charts/patches that perform better ?

Other possiblities, but at relatively much lower consistency than Laboratory grade references, would be to use high quality paint colors like Schmincke HKS Designers' Gouache.

Cheers,
Bart
« Last Edit: April 05, 2011, 10:12:15 AM by BartvanderWolf » Logged
digitaldog
Sr. Member
****
Offline Offline

Posts: 9188



WWW
« Reply #3 on: April 05, 2011, 02:50:58 PM »
ReplyReply

From the ColorSync list (from the creator of the mentioned BabelColor White Target). Good luck!
Quote
Re: Spectrophotometer accuracy

Subject: Re: Spectrophotometer accuracy
From: <email@hidden>
Date: Thu, 25 Aug 2005 16:34:18 -0400
Delivered-to: email@hidden
Delivered-to: email@hidden
While the article recommended by Ken is an excellent pointer for anyone wanting to improve inter-instrument accuracy, here are a few remarks based on the practical experience of two persons I know.

Both of them wanted to have an absolute standard, and both used the BCRA route. After spending thousands of dollars in material and services (one bought new, the other used, one of them spent near or over 10K in the project), they still cannot reach that goal.

The main reason is NIST traceable data is available, from the tile suppliers or from calibration centers, for a different spectrometer geometry than the one used by most graphic professionnals. The Eye-One or Pulse use the 45/0 geometry while, just as an example, Avian Technology (One supplier of BCRA tiles) use an 8 degrees/hemispheric geometry. Of course, the measured spectrums are different with each geometry.

Also, you need a temperature stable setup to optimise the measurement as the tiles are thermochromic (cyan is a good choice as it minimizes this effect).

Here is an article which tackles these problems.
http://www3.interscience.wiley.com/cgi-bin/abstract/50248/ABSTRACT
-------------------------
Color Research & Application
Volume 22, Issue 1 , Pages 51 - 60
An abridged technique to diagnose spectrophotometric errors
Roy S. Berns *, Lisa Reniff
Munsell Color Science Laboratory
Abstract
The effect on colorimetric accuracy of spectrophotometric errors (±0.5 and ±1% reference white, ±0.25 and ±0.5% reference black, and ±0.5 and ±1 nm wavelength) was simulated for the BCRA Series II tile set by using previously investigated models for these errors. E*94 color differences of up to 5.2 could result. Because the errors are linearly related to CIELAB differences for each tile, a technique was derived that enabled L*, a*, and b* coordinates between a calibrated tile and its measured value to be transformed into estimates of reference white, reference black, and wavelength error. The cyan tile was identified as the most effective tile for this technique. Universal equations are included for 45/Normal and Normal/Total geometries traceable to the National Institute of Standards and Technology (NIST), and Normal/45, Normal/Total, and Normal/Diffuse geometries traceable to the National Physical Laboratory (NPL). Measuring the cyan tile on a regular basis and transforming its colorimetric coordinates into spectrophotometric error metrics provides a useful method to validate the accuracy and reproducibility of a color-measurement instrument.
---------------------------------
All of this is color science heavy and requires a lot of time. It does not mean it cannot be done but simply points to the fact that there is no simple point and click solution to the problem.

Danny Pascale

email@hidden
www.BabelColor.com
Logged

Andrew Rodney
Author “Color Management for Photographers”
http://digitaldog.net/
stefohl
Jr. Member
**
Offline Offline

Posts: 59


WWW
« Reply #4 on: April 05, 2011, 05:21:36 PM »
ReplyReply

But wouldn't a simple solution be to measure a ColorChecker card when the spectro is new, save those measurements and then remeasure the same chart after one year? Not an absolute solution, but enough to notice if the calibrator is drifting away from the results it gave when new, and hopefully accurate.

Stefan
Logged

Stefan Ohlsson
Projektor
www.projektorutbildning.se
BartvanderWolf
Sr. Member
****
Online Online

Posts: 3752


« Reply #5 on: April 05, 2011, 06:59:01 PM »
ReplyReply

But wouldn't a simple solution be to measure a ColorChecker card when the spectro is new, save those measurements and then remeasure the same chart after one year? Not an absolute solution, but enough to notice if the calibrator is drifting away from the results it gave when new, and hopefully accurate.

Hi Stefan,

Sure, with the same CC and spectro it is easier than with different types of spectros. However, the question remains which one is drifting, the CC or the spectro, or both. I suppose that careful (cool, dry, dark, environmentally sealed) storage of the CC will help to keep it more stable.

Cheers,
Bart
Logged
stefohl
Jr. Member
**
Offline Offline

Posts: 59


WWW
« Reply #6 on: April 06, 2011, 01:41:24 AM »
ReplyReply

I suppose that careful (cool, dry, dark, environmentally sealed) storage of the CC will help to keep it more stable.

Hello Bart.

I've measured my ColorChecker now and then. The chart has been stored dark and dry, but at room temperature. As I have several EyeOne spectros, it's easy to see if the problem is with the chart or the spectro. The chart doesn't seem to change that much, if stored with normal care.

Stefan
Logged

Stefan Ohlsson
Projektor
www.projektorutbildning.se
PhilipCummins
Full Member
***
Offline Offline

Posts: 103


« Reply #7 on: April 06, 2011, 01:49:03 AM »
ReplyReply

But wouldn't a simple solution be to measure a ColorChecker card when the spectro is new, save those measurements and then remeasure the same chart after one year?

Technically you wouldn't know if the chart or the spectro has drifted unless you measure a couple of them with different spectrophotometers to get a good idea about the properties of the card & spectrophotometer.

Mt Baker Research offers tiles for calibration purposes (quite expensive) and should be accurate if stored properly. X-Rite/GMB from memory used to offer tiles for their inter-device agreement software but switched to cards for the latest NetProfiler software which are replaced every year. I guess in theory purchasing some older tiles and measuring with a known spectrophotometer could set a standard for measuring drift in that device only.
Logged
Rhossydd
Sr. Member
****
Offline Offline

Posts: 1965


WWW
« Reply #8 on: April 06, 2011, 02:38:49 AM »
ReplyReply

Thanks for everyone's contributions so far. Some interesting food for thought.

I already have measured CC24s and am a bit wary of trusting them too much given their variability. I guess the interesting question is which is more stable the CC24 or the i! ?

Yes, Andrew I take you point that delving too deep into this can be frustratingly complex and potentially misleading, but I also had hoped that I had make myself clear that I'm not looking for absolute lab accuracy here. Just a warning if measurements are deviating away from where they should be and possibly some guidance as to how often spectros actually need to be sent for re-calibration.

Ernst: Have your tests seen any significant deviations in performance over time ?
Logged
Ernst Dinkla
Sr. Member
****
Offline Offline

Posts: 2867


« Reply #9 on: April 06, 2011, 06:54:03 AM »
ReplyReply

For the SpectroCam, yes. The reason why I bought the Eye 1 Basic. Which isn't a year old so hasn't been checked again. June it will be a year old.

I would expect 3mm thick cast acrylic, opaque color samples mounted on a black board to be more consistent in time than a Color Checker card.

The BabelColor White Target is basically a PTFE tile with improved optical qualities like Spectralon is an improved PTFE. For this purpose a piece of PTFE (Teflon) 2" diameter and 10-15 mm thick mounted on a black background would be sufficient. With that size you are at the maximum reflectance of the stuff, no increase above it. A physics researcher at the university here wrote me he uses plain Teflon that is waterproof sanded for each calibration of a spectrometer. They all have a higher white reflectance than the usual calibration white delivered with the spectrometers.
 
The calibration white delivered with the spectrometer is another part that can shift color in time. I suspect that was part of Spectrocam's shift but couldn't cross check. While you can check the spectrometer to some consistent samples in time if the OEM calibration tile deviated you are still off. The one of the Eye 1 basic looks a bit more professional though.

What I would like to measure or get a spectral plot for is the light of the Eye 1 Basic, UV enabled version. The RAW measurement, not the tweaked measurement that my Eye 1 delivers when I put it on for example Spectralon that gives a near 100% reflectance on all wave lengths. I don't think that information is available outside X-Rite premises.


met vriendelijke groeten, Ernst

New: Spectral plots of +250 inkjet papers:

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
Logged
eronald
Sr. Member
****
Offline Offline

Posts: 4118



« Reply #10 on: April 11, 2011, 04:06:00 PM »
ReplyReply

Lamps with known spectra?

E.
Logged
Ethan_Hansen
Full Member
***
Offline Offline

Posts: 114


WWW
« Reply #11 on: April 11, 2011, 06:53:17 PM »
ReplyReply

What I would like to measure or get a spectral plot for is the light of the Eye 1 Basic, UV enabled version. The RAW measurement, not the tweaked measurement that my Eye 1 delivers when I put it on for example Spectralon that gives a near 100% reflectance on all wave lengths. I don't think that information is available outside X-Rite premises.
You can get the raw readings by using Argyll's "high resolution" mode. This provides the raw measurements at 3.3nm intervals rather than the massaged 10nm data.

We use the following methodology for tracking shifts in our fleet of spectrophotometers. It may not be perfect, and it does not give absolute accuracy values, but it certainly allows diagnosing when a particular instrument is out of calibration. For background, we have half a dozen Spectrolino/Spectroscans, 4 iCColors, and i1iSis automated readers. The other spectros (i1, DTP24, etc.) are not used often enough to worry about. It became readily apparent that we needed a method of tracking instrument drift and when servicing was required.

Our first approach was to use sets of plastic tiles. A visit to a big-box home supply store with a DTP24 in hand revealed several neutral solid-surface counter top materials that had reasonably flat spectral response curves. None were available in bulk, but the manufacturers were happy to supply samples. We measured these on the Spectrolinos and then measured small test charts formatted for the iCColor scanners printed on OBA-free paper. The plastic tiles had minimal temperature color shift (within Spectrolino measurement noise) over a +/-2C range. The small charts were reprinted regularly at a local Silver-Halide lab, and we compared Spectroscan readings vs. iCColor for those. The normal failure modes, in order of occurrence frequency, were: instrument Calibration reference being dirty, Spectrolino measurement filter collecting dust, and the measurement lamp getting too dim. We could detect these problems when they arose, but did not have enough resolution to track performance degradation.

Our next approach was to use a set of 12 BCRA tiles. We picked on up for not too terribly much on fleabay. Absolute measurement accuracy was not something we could determine, primarily because of the difference in the measurement geometry between the reference values and the 45/0 geometry all our spectros use. Following the research reported on above, the cyan tile did indeed show the best correlation both between our Spectrolinos and the values GMB and X-Rite gave when the instruments were sent in for calibration. (A single cyan tile costs $224 - not that bad. We later had the opportunity to have our tiles measured by a reference-grade, 45/0 instrument, but that had no real bearing on instrument tracking. We used the same methodology as before for tracking shifts in the spectrophotometers that could not measure the tiles; i.e. read a chart on the 'Linos and then remeasure on the other equipment.

The methodology we settled on for the last couple of years goes back to our original plastic tile strategy. The Spectrolinos still are checked using the BCRA tiles. We then use a Teflon sheet (this appears to be the one) to measure the white response of the iCColor and iSis spectros. It took a fair amount of swearing to stencil the necessary positioning markers on the Teflon using a black marker, but the process is simple: Measure the Teflon reference on the Spectroscan, averaging many points together, measure "patches" (all white) on the other instrument, average and normalize to the Spectro results, and trend the values. With regular instrument use, we can see the effects of iCColor lamp dimming over a period of several months. When the values are sufficiently out-of-whack, the instrument goes back to X-Rite for repair. So far, we have yet to wear out an iSis -- the LED lamps do indeed have a longer life.

The Teflon sheets are remarkably stable. We bought a pack of them - 10 or 12 - and have replaced them when the surface gets damaged. We measured each sheet on Spectroscans initially, using 1000 samples per sheet. After a couple of years of dark storage, the sheets have not shifted color in relation to either a freshly factory-calibrated instrument or the BCRA tile set by more than the Spectrolino accuracy.
Logged

Ernst Dinkla
Sr. Member
****
Offline Offline

Posts: 2867


« Reply #12 on: April 12, 2011, 01:54:17 AM »
ReplyReply


You can get the raw readings by using Argyll's "high resolution" mode. This provides the raw measurements at 3.3nm intervals rather than the massaged 10nm data.


Ethan,  thank you.   Yes, it is time I must embrace Argyll fully.

Your spectrometer testing methods assures me I was not on the wrong path. Have to rely on one though instead of many but I guess that makes life simpler too.


met vriendelijke groeten, Ernst


New: Spectral plots of +250 inkjet papers:

http://www.pigment-print.com/spectralplots/spectrumviz_1.htm
Logged
EsbenHR
Newbie
*
Offline Offline

Posts: 39


« Reply #13 on: April 12, 2011, 04:43:49 AM »
ReplyReply

Hi - this topic is also important for some of my own work.

It is rather difficult/expensive to test the absolute accuracy of a spectrometer.

However, a simple and cheap way to check the meter is to measure two (or more) filters on top of a stable light source.
When you stack the filters, the resulting absorption (for each wavelength) should be the absorption of the filters multiplied.

This assumes that you can get the spectrum itself. I use the following utility from Argyll with an XRite i1:

spotread -H -s logfile.log

The result is pretty easy to read. I imagine Excel would suffice for this purpose.

One of the best gray patches I have found are the neutral patches in the creative enhancement patches on the XRite passport.
They are actually more neutral than the big grey card (and a huge improvement over the gray patches on the color checker).


Hope this helps.
Logged
Greg_E
Full Member
***
Offline Offline

Posts: 113


« Reply #14 on: April 13, 2011, 09:00:24 AM »
ReplyReply

Has anyone checked white and black Acetal/Delrin plastics? Both seem to be fairly stable but I haven't done the kind of testing that Ethan or Ernst has done. The white Delrin did make a great backing device for measuring targets since it showed no OBA and was fairly flat across the spectrum.

Also FYI, the calibration tile in the Pulse is ceramic (metal ceramic is my memory is correct), I think the i1 uses some sort of plastic tile (but I cold be wrong).
Logged
Pages: [1]   Top of Page
Print
Jump to:  

Ad
Ad
Ad