Calman question...DeltaE 1976 vs 1991 vs ...?

They all tell you slightly different information, so it depends on what you want to know.

If you just want to know the absolute perceptual error from the target I'd go straight to dE2000. It's the most uniform for error reporting as a single number.

dE94 lets you break out the error as Chroma (saturation), Hue and luminance which makes CMS much easier to calibrate.

dE76 is older and is used in many standards and in CalMAN by default we factor the luminance out for grayscale. So if you don't care about gamma and just want to get white point correct, this can be helpful for that.

No. dEuv for grayscale and dE94 for CMS (color).

Actually, the difference between these formulas are not small. They can be quite dramatic. Add to that the fact that they scale differently and have different implications for color and white point, and have different results for different colors, it is important to know how to interpret the data.

Assuming a Rec. 709 gamut, consider an oversaturated green color of x0.298, y0.678, Y0.7152. CIELAB reports 39.9, CIELUV reports 22.6 and CIE94 reports 7.7. These are not "slight" differences.

Another example. Try to fix this color by lowering luminance to 0.512. CIELUV reports that the error decreases by half to 11.3. However, if you use CIE94 the error increases by nearly 40% to 12.4. These are huge differences.

I always recommend using CIE94 for color (you can use CIEDE2000 also, but the results are very similar) and try to get the error at 1.5 or below. Which formula you use for white is not critical. Just try to get the dE to 2.0 or lower. The reason for the difference is that CIE94 treats luminance different (and better) and luminance is not considered in grayscale at all.

Finally, you can get dH, dS, and dH data using the 1976 formulas just as easily as with the more recent formulas.

Right because gamma is irrelevant?

No. Gamma and grayscale tracking are, uh, different.

Funny I thought 50% gray was only one point.

If you are trying to make some point, please do so. This thread is about how to use various dE formulas for grading color accuracy. What does gamma have to to do with this?

It is an important component of overall data. When doing grayscale with any kind of multipoint control you should use a dE formula that incorporates luminance. Error in your gamma will manifest as saturation errors throughout the entire gamut. It's critical to minimize not just chromaticity errors, but also luminance errors.

Granted if you only have 2 points or 1 point, than luminance isn't actionable information.

One issue in this line of discussion is that I don't believe gamma measurements have ever been standardized, so the particular display and pattern affects measured brightness to some extent. My still-favorite example of "calibration stupidity" were people using 10-point gamma adjustments to get an adjusting-iris projector to measure "correctly". They thought getting their adjusting-iris display to make a "good" gamma graph meant they were "calibrating" gamma. Anyway, it just seems to me that currently gamma is somewhat of an inexact target, in spite of how it can affect an on-screen image.

Then could you explain this comment at the beginning of this thread: "by default we factor the luminance out for grayscale."


You seem to switch here from grayscale to color gamut measurements. Errors in the gamut require luminance data. It just isn't the same for grayscale, as the quote I have provided here seems to acknowledge. Why else would you ignore it by default? Luminance errors in the grayscale are treated separately as gamma, which has nothing to do with dE.

I didn't say I necessarily agreed with that, just that it's been the defacto industry standard for some time so that's the way our workflows operate out of the box.


I wouldn't ignore it by default, but then again you only want to present your users with actionable data, so there is a bit of give and take there with the learning curve.

The grayscale being the basis of the gamut, it's curve effects the location any color that isn't 100% saturated. If gray 25% is off luminance any triplet that uses a component at 25% will be off hue, off luminance and off saturation. Luminance error in grayscale don't just make your grayscale off, they throw off your entire gamut.

Do you think the researchers that came up with the dE formulas cared one lick if the value was "grayscale" or "gamut" those are calibrator terms dealing with displays. The people that created the dE formulas are working on appearance models and our ability to perceive a difference between two colors. If 75% gray has a dE2000 of 4.2, but it's exactly d65 it would look noticeably different than what the target 75% gray would be.

This is very true. It's also why where good standards exist the specify the entire viewing environment. The ambient light, the display surround, and the absolute black and white levels of the display will all effect your perception of gamma.

There is a proposed gamma standard for HDTV viewing called BT.1886, but it still doesn't' fully specify the viewing environment. It does make some good advancements such as including the displays black level in the gamma formula to address how contrast ratio effects perception.

True that and for a very good reason. Were it otherwise, very strange consequences would ensue that would make the calibration process very difficult. For example, the same color of white could have very different dE values depending on its level of stimulus. Also, the measured RGB balance could be very much off at low light levels and still show an acceptable dE value.


The problem with this is what is "off luminance?" Just considering what you write below, although the colors of a specified gamut have very specific luminance targets, in contrast the correct luminance values in the grayscale depend upon the chosen gamma curve, and that will vary depending on the display's contrast ratio, viewing environment, and standard used in the mastering of the source material. There just is no one "correct" gamma. SMPTE-C, for example, has no specified gamma at all.

Since i am calibrating my projector now ..this is timely ... can i assume that i should turn off all "dynamic, adaptive ..iris/contrast settings" and get the gamma flat with the display's inherent native contrast ?

after calibrating ..i can then see if i like the look of any dynamic contrast settings and then turn them back on ?

If the whole point of dE is that the values below a threshold are not discernibly different, how could a value below the threshold be visually wrong?

Dynamic backlight and adjusting iris features often make meaningless gamma measurements from windows or fields. If you are able to disable or bypass such features it typically will not significantly change the actual on-screen gamma, but disabling those features will often make a very significant change in the gamma measurements of windows or fields. It sounds like you have the general idea.

The ugly stepchild for dE is dEuv... because uvL color space is more perceptually relevant than other spaces, the dE calculations aren't convoluted by correction factors that try to compensate for perceptual issues with the the color space itself. The correction factors built-in to the dE calculation tend to improve some things, but other things can be "off". Since we are typically only measuring shades of gray, primaries, and complimentary colors... it doesn't freakin matter if the dE calculation produces a large error at some color of green or mauve or puce if we never measure that color (which we aren't likely to do as calibrators of video displays). And it doesn't much matter if the dE calculation is "off" when you are measuring an inaccurate green or magenta if the intention (and capability of the display or video processor) is to make magenta accurate and the dE calculation is accurate for colors CLOSE to the point we are measuring. dEuv (and uvL space) gets rid of all the playing around the other dE calculation options do because they are trying to be perceptually accurate for all of the billions of colors that can be measured in 24-bit color space. For us calibrators... we only need a dE that's pretty accurate for the specific colors and grays we measure... that's 10-21 grayscale steps, 3 primaries, and 3 complimentaries plus some space around each of those points. We really don't care (or shouldn't care) if the dE calculation chosen is way off at some color we will never measure.

I agree entirely. So the question is how do we calculate dE for grayscale?

If you include luminance data in those calculations, low stimulus values will result in low dEs, even though the color error is clearly visible.

To take just one example, consider the following: Assume a 2.22 target gamma, 20% stimulus, 35 ft-L peak output, no gamma error, and the following color x0.313, y0.345. This is a very greenish gray, the error easily visible.

If you ignore the level of stimulus, then the dE value (CIELUV)

Reference Luv: L100, u0, v0
Measured Luv: L100, u-7.317, v10.85

is 13.1 and the RGB balance is
R91%
G104%
B89%

This is what you would expect.

However, if you include luminance information in the dE calculation in the way I think you have suggested, then even though the RGB balance is the same and the visible error is the same, the dE value is now 2.9. This is clearly wrong.

Reference Luv: L18.11, u0, v0
Measured Luv: L18.11, u-1.325, v1.96

Actually I'd assert that the dE formula is much more accurate when you use luminance.

If your monitor is sRGB calibrated these swatches will be at the as advertised levels.


I would say the dE formula is fairly accurate.

At luminance of 98.5 L you get dE of 13.5 76 and 94 and 11.6 in 2000
At L 18 you get dE 76 and 94 of 2.0 and dE 2000 of 2.3

Luminance is a huge factor in grayscale.

I have found this combination to work the best. For gamma, I don't use dE, rather just try to get it as close to target at each step as possible. I am able to get in within less than +/-0.01 on my LG LCD thanks to it's excellent 10-pt white balance controls (R,G,B controls at each level). One thing that eludes me is why LG still refers to the grayscale steps as 'IRE' and not % stimulus. Also, it has controls at 0% but they don't do anything at all.

If your LCDs are OFF at 0% what's there to do? The only light coming from the panel at 0% is the backlight leakage, the crystals are opaque.
I just spent the night messing with my LG testing Calman 5 (closed beta testing) and if you notice, a 10 point gray scale, starts at 10% not 0. 0 is only used for gamma.
Why is it there.. cause the guy writing the firmware in the control put it in. Just because it is there, doesn't mean it works or needs to work.

When did CalMAN 5 come out?

I hasn't, it is in closed beta

the blacks on my S-IPS panel are bluish, so it could help if it actually worked; maybe your VA panel is different?

Must be, 0 does nothing at all, 0% window is not visible against the area that is not part of the video. The entire panel is the same brightness of the backlight leakage. Since I use this in a bright kitchen with 2 window and a half lite door, backlight is set to 80 and there is no dark environment viewing.

We are off topic here, I will pm you with some other things I found with the CMS

ok

Still in closed beta but just about anyone willing to sign a NDA can get in on the fun or wait for open beta.