
Page 1 of 2  1  2 

Thread Tools 
I also tried changing CP to dE2000 and changing CalMAN to dEXXXX_LuminanceCompensated, without much change in difference.
Which result should be believed, and why such radical differences?
touring calibrator
Jeti 1211
Latest reviews:
Panasonic TCP50ST60, Samsung PN64F8500, Panasonic TCP60VT60, Sharp LC80LE857U, Insignia NS65D260A13, Sharp LC70UD1U, LG 55EA9800
Sponsored Links  
Advertisement 

When I was switching different options, if I remember right with dE2000 selected in CP the two programs matched at 100% and then diverged as it went down, with CM giving much lower readings at the low end.
Interestingly, here with dE1994, even at 100% there's a big difference.
touring calibrator
Jeti 1211
Latest reviews:
Panasonic TCP50ST60, Samsung PN64F8500, Panasonic TCP60VT60, Sharp LC80LE857U, Insignia NS65D260A13, Sharp LC70UD1U, LG 55EA9800
touring calibrator
Jeti 1211
Latest reviews:
Panasonic TCP50ST60, Samsung PN64F8500, Panasonic TCP60VT60, Sharp LC80LE857U, Insignia NS65D260A13, Sharp LC70UD1U, LG 55EA9800
Looking more carefully at the data and the settings in HCFR, it appears that the contributing factor which make the result about the same as CalMAN is more the selection of dE2000 instead of dE1994. The dE result with or without considering gamma (absolute or relative Y) makes little if any difference.
Larry
The original, prerevival version of HCFR even had an option to 'Include Y (Luminance) in dE' calculations. IOW, you could base dE solely on x,y or on x,y,Y. I'm guessing something similar is going on here.
I measured an 11 point grayscale run on an LG LED with CalMAN, then ChromaPure. The differences between the x, y, Y readings are small, yet the differences in dE are enormous. I overlaid the two programs so we can compare the numbers.
I also tried changing CP to dE2000 and changing CalMAN to dEXXXX_LuminanceCompensated, without much change in difference.
Which result should be believed, and why such radical differences?
1. Go to http://www.brucelindbloom.com/index.html?ColorCalculator.html
2. In a separate tab open http://www.brucelindbloom.com/index.html?ColorDifferenceCalc.html
3. In the Color Calculator select the following options, Ref white D65 and RGB model sRGB.
4. Type the x,y values in the provided spaces.
5. Type 1.0 for Y.
6. Click the xyY button. This populates all of the other color spaces with equivalent values.
7. Go to the Color Difference Calculator.
8. In the Lab Reference type 100, 0, 0
9. In the Lab Sample type the Lab equivalents to the xyY values shown in the Color Calculator.
10. Click Calculate.
Results will display in CIELAB, CIE94 (Graphic Arts and Textiles), CIEDE2000, and two versions of CMC
Example
The colorimetric values you provided at 30% are 0.275, 0.281, 1.0
The Lab equivalent is 100, 4.893, 26.4308
The CIE94 value for this color is 26.9
There was a time when CalMan's default method for calculating grayscale dE included luminance, which is generally accounted for by gamma. The result is that the exact same color would have substantially decreasing dE values as you went lower on the video scale. However, this appears to not be the case here because the large discrepancy exists even at 100% video.
Tom Huffman
ChromaPure Software/AccuPel Video Signal Generators
ISF/THX Calibrations
Springfield, MO
There was a time when CalMan's default method for calculating grayscale dE included luminance, which is generally accounted for by gamma. The result is that the exact same color would have substantially decreasing dE values as you went lower on the video scale. However, this appears to not be the case here because the large discrepancy exists even at 100% video.
I just checked the dE charts in design mode for the Quick Analysis and HT Enthusiast workflows in Calman 5.3, and each has "Include Luminance Error" checked in its Properties.
This thread talks about it a little more, but I'm still unclear as to when and why you would want to include or exclude luminance from the calculation (same for profiling meters):
http://www.spectracal.com/forum/viewtopic.php?f=94&t=4150
I just checked the dE charts in design mode for the Quick Analysis and HT Enthusiast workflows in Calman 5.3, and each has "Include Luminance Error" checked in its Properties.
This thread talks about it a little more, but I'm still unclear as to when and why you would want to include or exclude luminance from the calculation (same for profiling meters):
http://www.spectracal.com/forum/viewtopic.php?f=94&t=4150
The thing is, all else is not equal. Film is a relatively dark medium and a white balance error at 2030% will almost certainly have more visual impact on what you see than an error at 90% where there is almost no actual content. Also, this approach is at odds with what we normally do when adjusting white balance using RGB indicators. The target is R100%, G100%, B100%. However, these are valid targets only if you assume a fixed level of luminance at 100%.
Regarding meter profiling, you would only include luminance if you had some reason to believe that the colorimeter being profiled was returning inaccurate luminance readings. Typically, they don't. Luminance is really easy, even for inexpensive colorimeters.
Lastly, this is not really relevant to Chad's original question, because as I and others have pointed out, the discrepancyat least with CIE94exists even at 100% video.
Tom Huffman
ChromaPure Software/AccuPel Video Signal Generators
ISF/THX Calibrations
Springfield, MO
You'd have to ask SpectraCal about the grayscale. I'm really not sure. I think that they would argue that, all else being equal, a colorimetric error at a high level of luminance is more perceptually evident than an equal error at a low level of luminance. But I hate to speak for them and they can certainly explain their approach better than I can.
The thing is, all else is not equal. Film is a relatively dark medium and a white balance error at 2030% will almost certainly have more visual impact on what you see than an error at 90% where there is almost no actual content.
OK, so including luminance errors can muddy the waters WRT white balance, which you wouldn't want if you prefer dEs to describe only white balance. OTOH, if you can get luminance close to correct, it won't contribute much, and then dEs do a better job capturing the whole error.
When I look at white balance, I try to get RGB to balance out at each IRE level 10%, 20%, etc. I don't see how to relate that to what you're saying. That is, I don't get what the issue is that you're referring to.
Tom Huffman
ChromaPure Software/AccuPel Video Signal Generators
ISF/THX Calibrations
Springfield, MO
What he means is that typical adjustments of white balance don't include any indication of how far off you are from your Y luminance target. CalMAN includes this information in their RGB graph by showing the three colored lines to be above or below the target line even though they might be "balanced".
What he means is that typical adjustments of white balance don't include any indication of how far off you are from your Y luminance target. CalMAN includes this information in their RGB graph by showing the three colored lines to be above or below the target line even though they might be "balanced".
Thanks, I was just about to post the following. Does it sound right?
What do the percentages refer to? In Calman, each IRE level is the same. There is a target luminance level and an RGB graph with +/ integer scale centered on zero, corresponding to the target level, and I try to get the RGB bars as close to zero as possible. Actually, I try to get the red and blue bars equal to green, and then I adjust gamma to try and get them down to zero as a group. I typically have to go back and forth a couple of times. Do your percentages correspond to the zero point in this graph and the bar lengths?
It's possible that you've unearthed an actual bug in the CM dE94 calculation. Given that CM, CP and HCFR appear to be in agreement on dE2000, I would lean in that direction.
Lot's of lines of code ... stuff happens.
This is exactly the case. It wasn't any sort of ideological choice. It was a bug and it's fixed.
SpectraCal
CalMAN Lead Developer
In what version is it fixed ?
What dE does allow for is calculating an error (I believe) for 1, 2, or 3 color coordinates. Now, there's no good use I can think of for calculating dE based on just x or just y or just u or just v.... but I think you could if it was something you needed for some kind of specialized work. For calibration, you COULD use just 2 coordinates for dE and get a useful result. In general, you would omit Luminance (Y or L) from the calculation. You can also calculate dEs for all 3 coordinates and that could RADICALLY change the dE result.
Because graphs are easier to deal with when they are 2D graphs, it MAY be easier for calibration use to use the two colorrelated coordinates for dE IF you are looking at a second graph to make sure you have your third coordinate (luminance) under control also. For grayscale, your gamma graph would suffice. For CMS you could do a 2D dE for color and use the color luminance graph to get that 3rd coordinate accurate. You NEVER want to compare a 2D dE calculation to a 3D dE calculation because you're going to get all messed up.
At one time (have not looked at this in a LONG time in CalMAN, could be very different now) you could select the dE you wanted to use for various dE calculations.... for grayscale you could pick xy or xyY to calculate dE. For color you had similar choices. I use dEuv for product reviews, but we COULD use dEuvL to get a 3D dE calculation. If you had red coordinates perfect, to the point that dEuv was ZERO, but there was a significant luminance error, your dEuvL for the same red point could be 5 or 10 or 25 or even 50 if the luminance error was large.
So while dE is useful, it is ONLY useful if you know what coordinates are included or not included so you understand what the error represents. You tell me dE1994 and I'm also going to want to know if you are using xy or xyY or uv or uvL or whatever other coordinates you may have decided on. When 2 people are discussing dE if one of them is using 2D dE calculations and the other is using 3D dE calculations, they are going to have a HELL of a hard time agreeing on anything!
The talk about the possibility that some software is "weighting" dE calculations... that sounds like nonsense to me because... the formulas for calculating dE don't allow for it. If you happen to be using dE 1994... there is just one way to calculate dE, you can't put your own "spin" on it because if you do, it's no longer dE1994... it's something else.
What dE does allow for is calculating an error (I believe) for 1, 2, or 3 color coordinates. Now, there's no good use I can think of for calculating dE based on just x or just y or just u or just v.... but I think you could if it was something you needed for some kind of specialized work. For calibration, you COULD use just 2 coordinates for dE and get a useful result. In general, you would omit Luminance (Y or L) from the calculation. You can also calculate dEs for all 3 coordinates and that could RADICALLY change the dE result.
Because graphs are easier to deal with when they are 2D graphs, it MAY be easier for calibration use to use the two colorrelated coordinates for dE IF you are looking at a second graph to make sure you have your third coordinate (luminance) under control also. For grayscale, your gamma graph would suffice. For CMS you could do a 2D dE for color and use the color luminance graph to get that 3rd coordinate accurate. You NEVER want to compare a 2D dE calculation to a 3D dE calculation because you're going to get all messed up.
At one time (have not looked at this in a LONG time in CalMAN, could be very different now) you could select the dE you wanted to use for various dE calculations.... for grayscale you could pick xy or xyY to calculate dE. For color you had similar choices. I use dEuv for product reviews, but we COULD use dEuvL to get a 3D dE calculation. If you had red coordinates perfect, to the point that dEuv was ZERO, but there was a significant luminance error, your dEuvL for the same red point could be 5 or 10 or 25 or even 50 if the luminance error was large.
So while dE is useful, it is ONLY useful if you know what coordinates are included or not included so you understand what the error represents. You tell me dE1994 and I'm also going to want to know if you are using xy or xyY or uv or uvL or whatever other coordinates you may have decided on. When 2 people are discussing dE if one of them is using 2D dE calculations and the other is using 3D dE calculations, they are going to have a HELL of a hard time agreeing on anything!
1000 x
profiling & calibration workflow tools: Display Calibration Tools
meter: Klein K10 A, i1Pro, i1D3
AVS thread: Lightspace & Custom Color Patch Set & Gamma Calibration on Panasonic 65VT60
The talk about the possibility that some software is "weighting" dE calculations... that sounds like nonsense to me because... the formulas for calculating dE don't allow for it. If you happen to be using dE 1994... there is just one way to calculate dE, you can't put your own "spin" on it because if you do, it's no longer dE1994... it's something else.
You can choose to calculate dE based solely in two dimensional space (x,y) or three dimensional space (x,y,Y). The latter option effectively "weights" the calculation so that rather large (x,y) errors are masked by the much smaller differences in Y (from the target point.) I've seen (x,y) dE errors of 86 or more at 10% stimulus drop to 69 (ish) once Y is taken into account.
Clearly one method (x,y) is probably more useful from a calibrators perspective. Depends on whether you're just concentrating on hitting D65 or if you're trying to do gamma/EOTF at the same time.
Whether large (x,y) errors at low luminance levels are really worth "sweating over" is another discussion.
As always, YMMV, which is why it's nice to have the option to choose between the two methods.
PS: To be honest, I don't really pay any attention to dE until after I'm done "twisting the knobs." I don't find it all that useful from a "mechanical" viewpoint.
You can choose to calculate dE based solely in two dimensional space (x,y) or three dimensional space (x,y,Y). The latter option effectively "weights" the calculation so that rather large (x,y) errors are masked by the much smaller differences in Y (from the target point.) I've seen (x,y) dE errors of 86 or more at 10% stimulus drop to 69 (ish) once Y is taken into account. .
There is ZERO "weighting" of dE by including or excluding luminance from the calculation. You are simply doing two entirely different calculations. Since there is NO luminance included when you use xy or uv or any 2 other coordinates, there's no "weighting" when you add luminance.... you simply add luminance to the calculation. Much like air speed vs land speed.... air speed includes the effects of headwinds or tail winds  this is not "weighting" either, it's simply 2 different ways of measuring speed  each has it's purposes... as do dE calculations with or without luminance included.
One of the things that has improved to some extent with each newer method of calculating dE actually involves "weighting" to make some errors that had small error values in prior calculation methods but were fairly easy to see, es[ecially in sidebyside comparisons. In this case "weighting" does apply because newer formulas try to avoid missing or understating some errors so that ALL errors are equally represented. We still not quite there yet, but dE calculations, in general, are getting better with each new generation of adopted calculation methods. This is ENTIRELY different than using or not using luminance in the calculation. The changes in dE formulas attempt to make a error in, say, salmon pink/orange have the same dE number as an equivalent error (equivalent in how visible the error is) in, say, green, or red, or some other common color.
Uhhhhh, YMMV doesn't apply to dE calculations. You can use 1, 2 or 3 coordinates in the calculation. If 2 people measure the same coordinates and both perform the SAME dE calculation using the same number of coordinates, their calculations will NOT vary... they will be the same. Which is the whole point of dE. As long as you are making accurate measurements, and using the formulas correctly and are using the same coordinates (usually 2 or 3 coordinates), you will always get the same result. Software like CalMAN allows you to select the coordinates you use for your dE calculations... as long as whoever you are talking to or comparing notes with is using the same coordinates and the same dE method of calculation (i.e. 1931, 1972, 1994, 2000, etc.) you will be making valid comparisons. (this of course assumes the software doesn't have a bug in dE calculations as CalMAN apparently did for some period of time).
Well, you SHOULD be paying attention to dE as you go along... using it as a sanity check can save a TON of time. I look at 3 dEs (usually 1931, uv, and 2000) as I make measurements because graphs don't always tell you how visible an error is, but dE WILL tell you that. I use 2 coordinates plus a 3rd graph with enough "resolution" (fixed vertical axis so as not to be fooled by the software adjusting the axis to make a large error appear small). The setup of the 3rd graph (for luminance) is critical though. There have even been times when I will use xy and xyY dE calculations or uv and uvL so I can see the effects of luminance errors EASILY  and that's one of the major advantages of a flexible software program like CalMAN. A raw data table can be setup to show you dEuv in one row and you can then have CalMAN display dEuvL on the very next line. uvL color space is pretty useful for calibration since you are looking at one of the more perceptually accurate color spaces because the coordinates are hue, saturation, and Luminance which also happen to be the controls in video displays that have CMS controls. And dEuvL doesn't change periodically while the formulas for calculating dE1931, 1972, 1994, 2000, etc. each have their own unique formulas that attempt to treat all errors equally... with the newer versions getting closer... but none of them are "perfect" yet. Which is why you want to look at more than one dE calculation while you are calibrating. I find myself trying to make the errors for all the dE calculations I look at less than 3 with luminance errors as small as I can get them. You can't always make 2 or 3 or 4 dE calculations all come out lower than 3, but many times you CAN get a better calibration by looking at multiple dE calculations.
Really ... we're going to have another argument over "semantics?"
I mean "weighted" in the sense that including Y in the dE calculation has the effect of greatly reducing (or "weighting") the error "number" at low luminance levels vs. high luminance. Period.
At some point, I may have time and energy to wade through the remainder of your post, but for now you've lost me at the first sentence.
This stuck out upon the quick scan through. No version of dE can tell you what to do about an error it can only tell you that there *is* an error and assign a relative number to it. Thus, it is not useful while you're doing the actual "knob" tweaking. OTOH, dE is useful to determine *where* one might need to do further "tweaking."
I've found it useful to have a realtime dE update while adjusting a display CMS which can hit it's x,y target or luminance but not both.
The talk about the possibility that some software is "weighting" dE calculations... that sounds like nonsense to me because... the formulas for calculating dE don't allow for it. If you happen to be using dE 1994... there is just one way to calculate dE, you can't put your own "spin" on it because if you do, it's no longer dE1994... it's something else.
What dE does allow for is calculating an error (I believe) for 1, 2, or 3 color coordinates. Now, there's no good use I can think of for calculating dE based on just x or just y or just u or just v.... but I think you could if it was something you needed for some kind of specialized work. For calibration, you COULD use just 2 coordinates for dE and get a useful result. In general, you would omit Luminance (Y or L) from the calculation. You can also calculate dEs for all 3 coordinates and that could RADICALLY change the dE result.
Because graphs are easier to deal with when they are 2D graphs, it MAY be easier for calibration use to use the two colorrelated coordinates for dE IF you are looking at a second graph to make sure you have your third coordinate (luminance) under control also. For grayscale, your gamma graph would suffice. For CMS you could do a 2D dE for color and use the color luminance graph to get that 3rd coordinate accurate. You NEVER want to compare a 2D dE calculation to a 3D dE calculation because you're going to get all messed up.
At one time (have not looked at this in a LONG time in CalMAN, could be very different now) you could select the dE you wanted to use for various dE calculations.... for grayscale you could pick xy or xyY to calculate dE. For color you had similar choices. I use dEuv for product reviews, but we COULD use dEuvL to get a 3D dE calculation. If you had red coordinates perfect, to the point that dEuv was ZERO, but there was a significant luminance error, your dEuvL for the same red point could be 5 or 10 or 25 or even 50 if the luminance error was large.
So while dE is useful, it is ONLY useful if you know what coordinates are included or not included so you understand what the error represents. You tell me dE1994 and I'm also going to want to know if you are using xy or xyY or uv or uvL or whatever other coordinates you may have decided on. When 2 people are discussing dE if one of them is using 2D dE calculations and the other is using 3D dE calculations, they are going to have a HELL of a hard time agreeing on anything!
what I was referring to was more in reference to RGB balance values in CM5 vs. CM4... In CM5, the RGB errors are the lower end appear to be much smaller than those at the top end whereas it was more linear with CM4
Ted was the one who posted about this a while back (can't remember exactly when, probably six months ago). If you can find his original posts, that would better state what I'm talking about.
Fair enough ... and probably the shortest translation of part of what I was trying to convey.
Still, because there's no "directionality" with dE, it's possible to create "double" the error if one were to rely upon it as "Truth." For instance, say you're doing a 10pt greyscale and you manage to get 30% down to a dE of 2.5 (skewed slightly red,) you call it "good enough" and move on to 20%, which you get down to dE of 3 (but skewed slightly blue.) Is that really "good?" Or, do you really have an overall greyscale skew of dE = 5.5 between 20% and 30% stims ... one that I suspect is large enough to be visible on a ramp (if perhaps not on live video.)
Granted the above is a "theoretical" and a "wise" calibrator would have probably crosschecked with other tools. Just saying, sometimes chasing down numbers isn't always the best approach. (Further granted that most 10pts allow you to get much closer than 2.5 at each point.)
I need a nap now.
For example, you can take a skin tones color checker reading, look at the dE1994 hue bar graph showing the results of all the patches, and determine if the hue, on average, is pointing towards yellow or magenta.
touring calibrator
Jeti 1211
Latest reviews:
Panasonic TCP50ST60, Samsung PN64F8500, Panasonic TCP60VT60, Sharp LC80LE857U, Insignia NS65D260A13, Sharp LC70UD1U, LG 55EA9800

Thread Tools  
Show Printable Version Show Printable Version
Email this Page Email this Page


Posting Rules  