Pages: [1]   Go Down

Author Topic: Profile Verification With RGB Patch Image?  (Read 1656 times)

Beau

  • Newbie
  • *
  • Offline Offline
  • Posts: 4
Profile Verification With RGB Patch Image?
« on: January 30, 2023, 10:33:48 am »

I would like to independently verify the results of my display profile but things are not working as expected. I thought this could be done by creating a reference image and reading the R, G, B values. While this seems to work for sRGB it does not for larger color spaces/profiles that correspond to my display specs (Display P3 or DCI P3). On a 0-255 scale the sRGB profile image varies by only about 1-2 so less than 1% the Display-pP3 profile image values are significantly different, data attached.

The first image is measuring sRGB patches and the second the results for sRGB and Display-P3.

Hoping someone can point to a flaw in my approach or recommend a better approach. Sorry for the lengthy post but wanted to be complete. Thanks for any help.

Test Images

Using Affinity2 (Photoshop style editor), I made a test image with various 0-255 R, G, B values. I “assigned” rather than converting to the output profiles to ensure the same values for each profile. I then exported as two 16-bit TIFFs with embedded sRGB and Display P3 profiles. I confirmed this with the Finder > Get Info which show the expected profiles for the two files.

•   Title: RgbPatches_SRGB
•   Color space: RGB
•   Color profile: sRGB IEC61966-2.1

•   Title: RgbPatches_DisplayP3
•   Color space: RGB
•   Color profile: Display P3

Image Viewer

I viewed the images using macOS Preview which I would expect is color managed, but was also able to “Assign Profile” to “sRGB IEC61966-2.1” or “Display P3” with no change in measurements.

Measurement

I used the macOS Digital Color Meter app with the respective “Display in sRGB” or “Display in P3” the app also displayed the correct profile name.

Computer & Display Specs/Setup

After encountering discrepancies on my second/external LG display I decided to first try with the built-in Apple display hardware.
•   2021 M1 MacBook Pro
•   XDR display
•   macOS 13.2 / Ventura
•   Preset Photography (P3-D65) This mode hides access to the settings (disables): Brightness, Automatically adjust brightness, True Tone, and Night Shift…
•   I have also disabled Battery & Energy > Slightly dim the display on battery.
I think this should provide the specified display gamut of "Wide color (P3)” or Display-P3?

I also have an external display LG 27” UltraFine 5K Display with specified Color Gamut (CIE 1931) DCI-P3 99%.

Any help appreciated.
Logged

Dave Ellis

  • Full Member
  • ***
  • Offline Offline
  • Posts: 103
Re: Profile Verification With RGB Patch Image?
« Reply #1 on: February 01, 2023, 12:06:12 am »

Reading RGB values in the way you have  is not going to achieve anything as the RGB values being read are effectively inputs to the display. To check display calibration etc, you need to measure the light output from the screen. You can do this with a colorimeter like you used to calibrate initially, in conjunction with appropriate software. DisplayCal would be a good choice as it  gives detailed measurement reports comparing standard color patch data and measured color patch data.

Dave
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 20708
  • Andrew Rodney
    • http://www.digitaldog.net/
Logged
http://www.digitaldog.net/
Author "Color Management for Photographers".

Beau

  • Newbie
  • *
  • Offline Offline
  • Posts: 4
Re: Profile Verification With RGB Patch Image?
« Reply #3 on: February 04, 2023, 06:33:59 pm »

Thanks for your responses, I will investigate. I am surprised that my x-rite ColorMunki Photo and now Calibrite software does not have some basic verification step to measure the "offsets", and maybe compare with the previous "offsets". From what I know human vision is pretty good for relative before-after or side-by-side measures but not absolute measurement. Thanks again.
Logged

digitaldog

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 20708
  • Andrew Rodney
    • http://www.digitaldog.net/
Re: Profile Verification With RGB Patch Image?
« Reply #4 on: February 04, 2023, 08:30:50 pm »

Thanks for your responses, I will investigate. I am surprised that my x-rite ColorMunki Photo and now Calibrite software does not have some basic verification step to measure the "offsets", and maybe compare with the previous "offsets". From what I know human vision is pretty good for relative before-after or side-by-side measures but not absolute measurement. Thanks again.
The verifications are usually just feel good approaches and not useful (I'd put the BabelColor as exception).

TThe validation is, by and large, only useful to determine when the process doesn't work correctly, and you want dE numbers to back it up. It isn't useful in gauging overall accuracy for a number of reasons. First and foremost is the issue is that you're using the same software and, more importantly, the same measuring instrument for both the calibration and validation, not a higher-grade reference device, to gauge accuracy.

As an analogy, suppose I measure my foot with a $2 wooden ruler from a hardware store and find it's 10.7 inches. Clearly, my foot isn't an accurate measuring device compared to that 1-foot wooden ruler. But how accurate is that $2 ruler? It, too, should be of sufficient accuracy for the task. There are devices like a 50-cent wooden ruler that may be accurate to +/- 1/10 of an inch. For measuring a 100-foot fence, that may be all that's needed in terms of measurement accuracy. For measuring a component that will be used on a spaceship headed to Mars, 1/10000 of an inch may just be adequate enough for the specifications of the job. But my 10.7-inch foot clearly isn't accurate. So when we talk about accuracy, we need to take into account the instrument used and the method of measuring the accuracy. And how accurate we need the results of the measurement to the reference. With dE, 1 or less is invisible. And the formula used, which differs, is also important. We are using the same instrument to evaluate the display accuracy, and that's not worthless but not ideal either.

Then we measure one area in the center of the display. What about each corner? Again, the dE reports provided by the software are probably useful when you see a really large dE (well over, say dE6). That could indicate something went wrong when you were calibrating/profiling. The screen saver came on, the software had an issue, the instrument fell off the screen, or some large amount of ambient light was measured. If the dE report is that high, you'd probably notice it anyway. So take the report with a grain of salt.

The other issue with some products providing such a report is they don't use too many color patches; they pick patches that are easy to produce and avoid those that could produce a lesser "ideal" report. Think dark saturated colors. And they often don't measure enough color patches to provide an average deltaE report for some.

Now what IS useful is known as colorimetric trending! That's where the same product reports the results over the course of time (say once a month over the course of a year). It tells you how often the display has changed if you calibrate every month and by how much in deltaE. It could be useful in deciding that instead of having to recalibrate every month, you can get by once a twice a month, or maybe you need to do this every few days. The inaccuracy of the device compared to a reference isn't a factor as long as the consistency in its measurements is good. And it should be, although not all available instruments are sadly.
Logged
http://www.digitaldog.net/
Author "Color Management for Photographers".

Wheathin21

  • Newbie
  • *
  • Offline Offline
  • Posts: 24
Re: Profile Verification With RGB Patch Image?
« Reply #5 on: February 05, 2023, 08:24:01 pm »

I would like to independently verify the results of my display profile but things are not working as expected. I thought this could be done by creating a reference image and reading the R, G, B values. While this seems to work for sRGB it does not for larger color spaces/profiles that correspond to my display specs (Display P3 or DCI P3). On a 0-255 scale the sRGB profile image varies by only about 1-2 so less than 1% the Display-pP3 profile image values are significantly different, data attached.

The first image is measuring sRGB patches and the second the results for sRGB and Display-P3.

Hoping someone can point to a flaw in my approach or recommend a better approach. Sorry for the lengthy post but wanted to be complete. Thanks for any help.

Test Images

Using Affinity2 (Photoshop style editor), I made a test image with various 0-255 R, G, B values. I “assigned” rather than converting to the output profiles to ensure the same values for each profile. I then exported as two 16-bit TIFFs with embedded sRGB and Display P3 profiles. I confirmed this with the Finder > Get Info which show the expected profiles for the two files.

•   Title: RgbPatches_SRGB
•   Color space: RGB
•   Color profile: sRGB IEC61966-2.1

•   Title: RgbPatches_DisplayP3
•   Color space: RGB
•   Color profile: Display P3

Image Viewer

I viewed the images using macOS Preview which I would expect is color managed, but was also able to “Assign Profile” to “sRGB IEC61966-2.1” or “Display P3” with no change in measurements.

Measurement

I used the macOS Digital Color Meter app with the respective “Display in sRGB” or “Display in P3” the app also displayed the correct profile name.

Computer & Display Specs/Setup

After encountering discrepancies on my second/external LG display I decided to first try with the built-in Apple display hardware.
•   2021 M1 MacBook Pro
•   XDR display
•   macOS 13.2 / Ventura
•   Preset Photography (P3-D65) This mode hides access to the settings (disables): Brightness, Automatically adjust brightness, True Tone, and Night Shift…
•   I have also disabled Battery & Energy > Slightly dim the display on battery.
I think this should provide the specified display gamut of "Wide color (P3)” or Display-P3?

I also have an external display LG 27” UltraFine 5K Display with specified Color Gamut (CIE 1931) DCI-P3 99%.

Any help appreciated.

As others have pointed out, this won't work. Displaycal has an excellent verification step after calibration and characterization, you can choose from a couple verification targets or generate your own. It makes a very nice and easy to understand report based on the verification target.
There is a pretty hard limit to how accurate your monitor is. Instruments have error, even the best spectrophotometers on the market can't do much better than .5 DeltaEs unless you do multiple measurements and average them. Colorimeters are even worse. Like others have said, it's a measure of if it worked or not, not how well it worked, or whether or not it's time to reprofile it.
Just my 2 cents.
Logged

aaron125

  • Jr. Member
  • **
  • Offline Offline
  • Posts: 55
Re: Profile Verification With RGB Patch Image?
« Reply #6 on: May 28, 2023, 05:31:53 am »

As others have pointed out, this won't work. Displaycal has an excellent verification step after calibration and characterization, you can choose from a couple verification targets or generate your own. It makes a very nice and easy to understand report based on the verification target.
There is a pretty hard limit to how accurate your monitor is. Instruments have error, even the best spectrophotometers on the market can't do much better than .5 DeltaEs unless you do multiple measurements and average them. Colorimeters are even worse. Like others have said, it's a measure of if it worked or not, not how well it worked, or whether or not it's time to reprofile it.
Just my 2 cents.
Not sure what brand and model of devices you’re referring to but most current consumer spectros, such as an i1pro2 or i1pro3 are far more accurate than the 0.5dE (76, 94, 2000?) you quoted. Eg, the i1pro3 has an inter-instrument agreement of 0.3dE00, far better than the 0.5dEyou quoted. When it comes to emissive measurements, it has an x,y repeatability of +/-0.002, and the i1d3 has repeatability of 0.001 x,y.

I regularly check my Eizo screen using PatchTool’s DisplayCheck feature, and over the 1617 patches, I often average 0.05-0.1dE00, according to my i1d3. Maybe 20+ years ago the devices might have been as inaccurate as the 0.5dE you mentioned but for at least 10-12 years or more, since the i1pro2 and i1d3 were released, they’ve been more than accurate enough for anyone using these consumer grade devices.

Just out of curiosity, exactly which device are you referring to, being so inaccurate at 0.5dE?
Logged

GWGill

  • Sr. Member
  • ****
  • Offline Offline
  • Posts: 609
  • Author of ArgyllCMS & ArgyllPRO ColorMeter
    • ArgyllCMS
Re: Profile Verification With RGB Patch Image?
« Reply #7 on: May 28, 2023, 08:54:38 pm »

Eg, the i1pro3 has an inter-instrument agreement of 0.3dE00, far better than the 0.5dEyou quoted. When it comes to emissive measurements, it has an x,y repeatability of +/-0.002, and the i1d3 has repeatability of 0.001 x,y.
i1Pro3 reflection inter-Instrument-Agreement is worst case 0.8 delta E 2K to the X-rite standard. So theoretically, two i1Pro3's could be 1.6 delta E apart (I hope that is unlikely in practice!). Agreement with instruments of different makes and models is likely to be worse.
The emission spec. doesn't say what the light source is (in the past it used to be an incandescent source), but it is a 5000K white light source, so not at all challenging, and it's hard to relate x,y 0.002 to dE.

From my own experience I know I'm pretty happy if I can get two consumer grade instruments to agree within 0.5 dE worst case across a range of color samples. It doesn't happen often.

Measuring more extreme samples can be quite a challenge. For instance attempting to measure a monochromatic LED light source with a fairly high end spectro I seemed to be seeing on the order of 0.2 dE temperature drift from the instrument (the LED being measured was temperature stabilised).
Quote
I regularly check my Eizo screen using PatchTool’s DisplayCheck feature, and over the 1617 patches, I often average 0.05-0.1dE00, according to my i1d3.
Average dE I believe you. Worst case, not so much, unless the same instrument was used for the actual calibration/profile.
Logged
Pages: [1]   Go Up