- This topic has 4 replies, 2 voices, and was last updated 5 years, 9 months ago by
.
-
Posts
-
Dear color management geeks,
I wonder if it makes sense to set the monitor brightness to 100% for creating a CCMX correction matrix for my I1 Display Pro. The Idea is to achieve the highest possible signal-to-noise ratio with my i1Pro2. Or is it possible, that the spectral curve changes when the brightness is reduced (apart from the intensity)? In this case, maybe it would be better to create the CCMX at the desired brightness of say 50%, despite a worse SNR?
Here in the forum I came across the X-Rite white paper “Display Calibration Devices: Methods, Accuracy, and Cost”.
It explains that colorimeters are more suitable for display calibration than consumer spectrophotometers, because the light becomes splitted in the spectrophotometer which finally leads to a worse SNR.
However, monitors do not have a continuous light spectrum but rather a narrow-band light spectrum. So there is not much to split, in my opinion.
Moreover a colorimeter has filters that only let a part of the light pass through. Thus because of the narrow-band spectrum each sensor of the colorimeter gets more or less only the red or the green or the blue component of the monitor spectrum. So the benefit is not as big as one might expect for a continuous spectrum.
The explanations in the white paper are absolutely plausible. But is the result really more accurate if you first create a CCMX correction matrix with the i1Pro2 and then calibrate with the I1Display Pro?
To be honest, I haven’t noticed any real differences yet. But unfortunately I have no possibility to test this with a laboratory device. What do you think? Has anyone tested it?
Florian, you wrote in another post: “…but you shouldn’t use a spectrometer for prolonged measurement sessions anyway if you can avoid it, as low light accuracy and drift are a real problem”. This corresponds to the white paper. Did you have the opportunity to compare the results of a colorimeter with the results of a consumer spectrophotometer with the help of a lab device?
I mean temperature drift should not be such an issue with the I1Pro2, is it?
Calibrite Display Pro HL on Amazon
Disclosure: As an Amazon Associate I earn from qualifying purchases.Dear color management geeks,
I wonder if it makes sense to set the monitor brightness to 100% for creating a CCMX correction matrix for my I1 Display Pro. The Idea is to achieve the highest possible signal-to-noise ratio with my i1Pro2. Or is it possible, that the spectral curve changes when the brightness is reduced (apart from the intensity)? In this case, maybe it would be better to create the CCMX at the desired brightness of say 50%, despite a worse SNR?
Here in the forum I came across the X-Rite white paper “Display Calibration Devices: Methods, Accuracy, and Cost”.
It explains that colorimeters are more suitable for display calibration than consumer spectrophotometers, because the light becomes splitted in the spectrophotometer which finally leads to a worse SNR.
However, monitors do not have a continuous light spectrum but rather a narrow-band light spectrum. So there is not much to split, in my opinion.
Moreover a colorimeter has filters that only let a part of the light pass through. Thus because of the narrow-band spectrum each sensor of the colorimeter gets more or less only the red or the green or the blue component of the monitor spectrum. So the benefit is not as big as one might expect for a continuous spectrum.
You can manage go get more light into the colorimeter with a big lens. All this light will go to just 4 sensors (i1d3) in a tiny area.
A graphic arts spectro has sensors spaced at 3nm. For a LED widegamut it’s about 100nm per emisive LED spectrum, you have Xrite corrections or DisplayCAL database to check out this. The drop in amount of light is huge if you compare those two devices. For dark colors the amount of light hiting spectrophotometer sensors is very very small compared with that kind of colorimeters. This is why they are slow.The explanations in the white paper are absolutely plausible. But is the result really more accurate if you first create a CCMX correction matrix with the i1Pro2 and then calibrate with the I1Display Pro?
i1Pro2+i1DisplayPro would be more accurate specially in dark colors…and faster as patch number gets higher (LUT3D creation).
If your display have very narrow spikes an i1Pro2 cannot even be able to measure them. It depends in nm value where those spikes happen but error could be huge or tiny if standard observer varies too much in that tiny nm interval (“ramps”) but “averaged” measure for 3nm step is not accurate or even “placed” wrong. ArgyllCMS programer did a test with something like CCFL backlight. I’m afraid that WGCCFL or PFS phorphor led will be much worse. You can do the maths and check by yourself and share here results since there is a huge DB fo CCSS corrections here for data mining to check against the same exact model from 1nm spectra (lab grade equipment??) from Xrite for a few displays (PA271W, PA242s, U2413, some iMac like…). Huge task but possible if you had the right amount of spare time.
I’d choose CCSS against that CCMX.
With CCSS you weight the same spectral distribution (10nm /3nm) against colorimeter observer (firmware 1nm data, which you can take as accurate as i1pro/i1studio nm spacing and SNR calibrated sensors) and standard observer. The same spectral distribution to correct observer differences. If the actual spectral distribution is “equal” (like an WLED), problem solved. If actual spectral distribution is more spiky at some nm and your colorimeter is almost equal to standard observer at that nm location, you measure that extra light, good! If actual spectral distribution is more spiky at some nm and your colorimeter is nor so close to standard observer at that nm location, it would be corrected by a certain amout in the correction. It’s a win-win but firmware data must be accurate and colorimeter filters should not fade “fast”. It seems to be the situation with an i1d3… very smart guys those who designed i1d3 device will those dichroic filters and colorimeter observer stored in firmware.
With CCMX you try to correct a limited spectral sample (10nm? 3nm?) weighted against standard observer versus actual spectral distribution weighted with colorimeter observer. Apples an oranges if you have spikes placed at certain nm.
In the end CCSS translates to a matrix when ArgyllCMS reads firmware observer. Try to use yor colorimeter ArgyllCMS command line with more verbose output with an without CCSS, then add some matrix multiplications/inversions to do by yourself since colorimeter raw data is in “counter” values that need to be traslated to XYZ coordinates by “counter to XYZ” matrix. So it’s possible to get a CCMX matrix for your particular colorimeter based on your own/community-made/vendor CCSS sample. Then port that matrix to whatever measurement software you like to check. CalMAN does nor support custom CCSS but you can put a matrix manually… and it works like a charm to port a CCSS correction to that inflexible and closed piece of software called CalMAN in this way.To be honest, I haven’t noticed any real differences yet. But unfortunately I have no possibility to test this with a laboratory device. What do you think? Has anyone tested it?
Very dark colors in a high contrast display. Spiky spectra, something like WGCCFL or a red measurement for P3 iMac. Try to measure them and compare, share here if you wish.
IMHO i1Pro2 + CCSS + i1DisplayPro is the way to go if you cannot know what kind of backlight is used in your display.
“Almost” the same price as an i1Pro2 alone, more ready to spiky spectra (3nm driver) like those P3 TVs/Displays, more accurate dark readings and faster, very fast.
I’ll get an i1d3 fist…if you need the graphics arts spectro add it later.Hello Vincent, thank you for your detailed comment. However, there is one point I do not understand.
You say:” With CCMX you try to correct a limited spectral sample weighted against standard observer versus actual spectral distribution weighted with colorimeter observer. Apples an oranges if you have spikes placed at certain nm.”
If I understand you correctly, you mean that I measure, exaggeratedly speaking, a “false” spectral distribution and weight it against a perfect mathematical model of the standard observer in the spectrophotometer on the one hand, and weight the actual (real) spectral distribution against a “false” colorimeter observer on the other hand.
I don’t see the disadvantage compared to CCSS.
In the case of CCSS you weight a “false” spectral distribution against the standard observer in the spectrophotometer and weight the same “false” spectral distribution against a false colorimeter observer.
Do you mean the advantage is that in the case of CCSS it is at least the same “wrong” spectral distribution?
Well, I’m not sure that’s an advantage. At least in my case, the delta E between the white point of the CCMX measurement and the spectrophotometer seems to be smaller than the one between CCSS and the spectrophotometer.
I have another general question.
If I use the self-created corrections (CCMX or CCSS), the xy values of the white point jump around in an interval of approx. 1 deltaE, to the preset white point, during measurement.
If I use corrections from other users, or the corrections provided by DisplayCAL, the displayed xy value is much more stable. The value does not change continuously. How could that be? A different correction matrix, even if it is wrong, does not suddenly make the measurement unstable?
That makes no sense to me.
[…]
I don’t see the disadvantage compared to CCSS.
[…]
Do you mean the advantage is that in the case of CCSS it is at least the same “wrong” spectral distribution?
Well, I’m not sure that’s an advantage.
First of all you should see what happens when 10nm or 3nm is not enough. Spikes get averaged value at nm measurement point.
-If colorimeter observer is not close to standard observer in that spike, it is “at least” corrected to averaged spike. That averaged value could be higher or lower that the light available in that “A nm – B nm” region. The actual amount of light in this spectral region will hit colorimeter which is at least corrected by averaged value.
-If colorimeter observer (firmware data) is very close to standard observer in that spike, it does not matter what kind of CCSS data you feed in that “A nm – B nm” region. When computing matrix correction translated from CCSS data, matrix components won’t carry correction from that region.As I said before, if you do not like this conjecture, you’ll have to do the maths. Time consuming but could shed light to your questions.
There are 1nm vs 10nm/3nm samples for well known displays from vendor or community data. Dump colorimeter observer from an i1d3. Then “measure” (computed).
Post here your results.For a valid empirical test youll need a higher end spectro like a JETI specbos and display with a very spiky spectral distribution:
-test CCMX obtained with an i1pro
-test CCSS obtained with an i1pro and with high end reference spectrophotometer.
Of course CCMX obtained from that high end reference spectrophotometer for that display and that “consumer” colorimeter will be the most accurate correction for that particular colorimeter.
Uncertainties about this CCMX vs CCSS issue arise because an i1Pro-family spectro is not what we may call a “reference” spectrophotometer to measure all kind of displays.Tests done with a i1pro and an i1d3 alone are not relevant, you need a reference to measure them (or different kind of corrections) against it. Unfortunately most people do not have access to that kind of reference instruments which make empirical test very difficult, to see these questions with your/our own eyes.
Different “spike” spectral distrubution will offer different results. For a worst case scenario you may want to look for spikes located where observer curves are not close to a local maximum. Local maximum means derivative close to zero in an interval close to that maximum, which means you could take a “near” constant averaged value for observer. It seems to be a situation more favorable for spikes averaged by a “lower but broader” spike in 10nm/3nm reading. If you wish to do the maths and post them here they would be helpful.
z-bar varies faster than the other observer curves. Errors from averaging (noise or diffraction grating) in displays using a blue LED may be observable between a high end spectro and an i1Pro.
I mean, I’m just pointing a few sources of errors and inacuracies if you wish to do a more detailed study.
At least in my case, the delta E between the white point of the CCMX measurement and the spectrophotometer seems to be smaller than the one between CCSS and the spectrophotometer.
This is the way CCMX are mean to work: to make a colorimeter measure “the same” that the spectrophotometer you take as “reference”. But we doubt that with certain spectral distributions a 10nm or 3nm measure could be called a “reference”.
This closeness bewteen measurements is not a proof, not even close, to what is better or worse. That’s why you need a better instrument to compare, and chosen spectral distribution will play a role here. I think that displays with spikes near a local maximum in standard observer will be more benevolent to CCMX in lower end spectrophotometers.
I have another general question.
If I use the self-created corrections (CCMX or CCSS), the xy values of the white point jump around in an interval of approx. 1 deltaE, to the preset white point, during measurement.
If I use corrections from other users, or the corrections provided by DisplayCAL, the displayed xy value is much more stable. The value does not change continuously. How could that be? A different correction matrix, even if it is wrong, does not suddenly make the measurement unstable?
That makes no sense to me.
It could be the display as it heats. IDNK what kind of display you are measuring. IDNK what CCMX you tested first.
Also if you take “raw” XYZ measures from a colorimeter and add them some noise, e(t), caused by display or by instrument, you get X+e, Y+e, Z+e. Let’s take noise equal in all coordinates for simplification. Multiply (left) by a 3×3 matrix correction you could see that different matrices will magnify that “e” in a different way in corrected XYZ values. A identity matrix in a CCMX will translate e unchanged.
Try to do the maths for your CCMX and another comunity CCMX and see which magnifies more an ex(t), ey(t) and ez(t) noise for each raw XYZ measurement and if that matches your findings, but keep in mind that the source of the noise and its variation with time (t) will play a role.Hello Vincent,
many thanks to you and your attempt to help me. I’m sorry that it takes so long for me to answer. But I don’t have much time right now. Moreover, I have to translate my thoughts into English first, this takes me a little bit of time, also 🙂
You say:
If colorimeter observer is not close to standard observer in that spike, it is “at least” corrected to averaged spike. That averaged value could be higher or lower that the light available in that “A nm – B nm” region.
The actual amount of light in this spectral region will hit colorimeter which is at least corrected by averaged value.
-If colorimeter observer (firmware data) is very close to standard observer in that spike, it does not matter what kind of CCSS data you feed in that “A nm – B nm” region. When computing matrix correction translated from CCSS data, matrix components won’t carry correction from that region.
So you mean that there are regions in the spectral distribution where the colorimeter observer is very similar to the standard observer. And in this regions the false measurement of a imprecise spectrophotometer does not harm, right?
Well, it makes sense to me, now. Thank you!
I also could imagine a perfect colorimeter with perfect filters which match the standard observer. So no correction is necessary. If we get an inexact spectral distribution from a 10nm spectrophotometer, the CCSS correction matrix becomes still an identity matrix. So no correction will be done, because the XYZ values from the spectrophotometer and the colorimeter will be the same, as the same (inexact) spectral distribution is used to compute the XYZ values.
If I create a CCMX for this perfect colorimeter, the correction matrix will not be an identity matrix, because the right XYZ-values from the colorimeter reading will not match the wrong XYZ-values from the spectrophotometer reading.
So as far as I can see, the accuracy of CCSS depends on how well the colorimeter observer matches the standard observer. If they match well, no correction will be applied by CCSS. That’s good.
If the match is low, there is probably no big difference between CCSS and CCMX. Because the reference XYZ-values from the spectrophotometer are wrong anyway.
Further investigation is required 🙂
It could be the display as it heats. IDNK what kind of display you are measuring. IDNK what CCMX you tested first.
It’s a NEC MultiSync P221W. It has a pretty wide gamut, even if it is not one of the SpectraView Displays. I do not think it has anything to do with the warmup. The display was already on for over an hour and the comparison measurements were made shortly after each other.
Also if you take “raw” XYZ measures from a colorimeter and add them some noise, e(t), caused by display or by instrument, you get X+e, Y+e, Z+e. Let’s take noise equal in all coordinates for simplification. Multiply (left) by a 3×3 matrix correction you could see that different matrices will magnify that “e” in a different way in corrected XYZ values.
OK, that seems possible to me. And again: Further investigation is required 🙂
Thank you Vincent, the discussion with you was very enlightening.
