I have a question that I need help with. I have a set that has desaturated red that I just cant push any farther. its close but not quite there yet. my delta E measures lower when reds value for Y, calculated from the Rec 709 chart, is higher than it is supposed to be. when I try to dial down the value to get it closer to where the correct value is, the delta E gets higher. I thought it would get lower. the color point on the cie diagram stays in the same position. whats going on here. according to delta E it would seem like I should adjust the gain for red up to compensate for its desaturation. is this correct. or should I ignore delta E and get that gain as close as possible. I hope i'm making sense.
Delta E or by the numbers. what to use.
-
Now Available: Tech Talk Podcast with Scott Wilkinson, Episode 19 Click here for details.
1 - 20 of 38 Posts
Joined
·
8,008 Posts
This is a complicated subject. The "correct" brightness for a color is an ambiguous target.
First, it can be correct relative to a specified gamut (e.g., Rec. 709) or correct relative to what is expected for a non-standard gamut. Those 2 numbers generally won't be the same. In particular, a non-standard gamut such as you describe with a desaturated red will expect a somewhat brighter red.
Second dE is an ambiguous term. There is not one, but several, dE formulas. Only the 1976 formulas (CIELUV and CIELAB) predict the type of reverse interaction between saturation and brightness you have noticed. Others do not.
This is a complicated subject. The "correct" brightness for a color is an ambiguous target.
First, it can be correct relative to a specified gamut (e.g., Rec. 709) or correct relative to what is expected for a non-standard gamut. Those 2 numbers generally won't be the same. In particular, a non-standard gamut such as you describe with a desaturated red will expect a somewhat brighter red.
Second dE is an ambiguous term. There is not one, but several, dE formulas. Only the 1976 formulas (CIELUV and CIELAB) predict the type of reverse interaction between saturation and brightness you have noticed. Others do not.
Is there some program or spreadsheet out there that recalculates the correct Luma targets (red's level versus white's measurement) to aim for based on your individual set's xy coordinates for the primaries?
My set has fairly oversaturated red and green vs. Rec 709's xy points. So if I am understanding your explanation above, if I adjust red to get 21% of white's measurement, I might be actually off in terms of correct Color setting for my set?
C.
Is there some program or spreadsheet out there that recalculates the correct Luma targets (red's level versus white's measurement) to aim for based on your individual set's xy coordinates for the primaries?
My set has fairly oversaturated red and green vs. Rec 709's xy points. So if I am understanding your explanation above, if I adjust red to get 21% of white's measurement, I might be actually off in terms of correct Color setting for my set?
C.
Joined
·
8,008 Posts
The math for this is published in the Dec. issue of Widescreen Review. I did post a spreadsheet here some time ago, but I don't recall where.
The math for this is published in the Dec. issue of Widescreen Review. I did post a spreadsheet here some time ago, but I don't recall where.
Joined
·
6,776 Posts
To answer your first question, C, yes there is. Both CalMAN and Progressive Labs' software package (CA-6X) offer the ability to do custom RGB transforms/color targets based on your set's actual primaries. This is very handy for sets like my old Mits CRT RPTV, which has no ability to adjust primaries. It's especially useful for knowing where the secondaries should be, since they are calculated from the primaries, and you won't be trying to hit secondary targets based on primary values that don't match what you have.
To answer your first question, C, yes there is. Both CalMAN and Progressive Labs' software package (CA-6X) offer the ability to do custom RGB transforms/color targets based on your set's actual primaries. This is very handy for sets like my old Mits CRT RPTV, which has no ability to adjust primaries. It's especially useful for knowing where the secondaries should be, since they are calculated from the primaries, and you won't be trying to hit secondary targets based on primary values that don't match what you have.
I am calibrationg to the rec709 standard and as far as I was aware the cie 1931 gaumat was what I was using. I am using color hcfr and it is set to rec709. to my understanding and what I see of the cie Diagram, I thought that color hcfr would automatically use cie 1931. if that is the case than it would seem that according to what you have said I should not be seeing the results I am describing. the blue on my set is consistantly desaturated on these sets.(probably an inaccurate color wheel). I have a Mitsubishi wd-65835. I am obviously missing something here. my sensor isn't the best (eyeone lt) but these results are repeatable. any other words of wisdom?
goto accupel.com and click on hdg4000 manual and download the display calculator. punch in the actual x y measurments of your primaries and it will tell you what the Y of your primaries should be and what the x y Y of your secondaries should be based on your primaries x y position.
goto accupel.com and click on hdg4000 manual and download the display calculator. punch in the actual x y measurments of your primaries and it will tell you what the Y of your primaries should be and what the x y Y of your secondaries should be based on your primaries x y position.
cool little program. it definitely will allow you to experiment but I'm not seeing where it will tell me anything that colorhcfr wouldn't. perhaps I'm not using it right yet. one question, is 50 ire gray the grey the program is expecting?
so we are all going on the assumption that the Y value should be different if the color in question is not the correct saturation for the color gamut. according to how I am understanding this, that isn't the case unless we are using cie 1974 but color hcfr is using cie 1931 and that is what I am using. this, of course still leaves the question of why I am getting bad delta E at close to the correct Y value.
so we are all going on the assumption that the Y value should be different if the color in question is not the correct saturation for the color gamut. according to how I am understanding this, that isn't the case unless we are using cie 1974 but color hcfr is using cie 1931 and that is what I am using. this, of course still leaves the question of why I am getting bad delta E at close to the correct Y value.
Joined
·
8,008 Posts
I think that you are confusing different gamuts (SMPTE-C vs. Rec. 709, for example) with different color spaces that depict those gamuts (1931 xy vs. 1976 u'v', for example). You can display the Rec. 709 gamut in either of these color spaces, though the 1976 version is more perceptually uniform.
BTW, an undersaturated blue on these displays is not uncommon.
I think that you are confusing different gamuts (SMPTE-C vs. Rec. 709, for example) with different color spaces that depict those gamuts (1931 xy vs. 1976 u'v', for example). You can display the Rec. 709 gamut in either of these color spaces, though the 1976 version is more perceptually uniform.
BTW, an undersaturated blue on these displays is not uncommon.
Joined
·
6,776 Posts
Yow! I forgot about Greg's cool little calculator. Sorry, guys.
![]()
Yow! I forgot about Greg's cool little calculator. Sorry, guys.
Joined
·
8,008 Posts
Me too.
Quote:
Originally Posted by Rolls-Royce /forum/post/15476326
Yow! I forgot about Greg's cool little calculator. Sorry, guys.![]()
Me too.
Joined
·
8,008 Posts
That is correct only for gray scale measurements. For pri/sec color the brightness of the color has a profound impact on dE.
That is correct only for gray scale measurements. For pri/sec color the brightness of the color has a profound impact on dE.
Joined
·
3,587 Posts
dE can be 2-D or 3-D and it "works" for gray or colors - any point within a color space. It is merely an expression of how "far" a measured point is from a reference/target point within any given color space. If the reference point is expressed in 2 dimensions (xy or uv most commonly), dE is only expressing that 2-D difference and L or Y could be way off. If all 3 dimensions are included in the measurement and reference point, dE can be a 3-D expression. A dEuv formula implies that this is only a 2-D calculation, which is not influenced by L or X. On the other hand a dEuvL calculation would require all 3 coordinates (for the target and measurement) and WOULD be influenced by L or X.
We (calibrators) don't typically USE dE with grayscale measurements, but we could if the software being used was setup that way. Before starting, you'd have to enter a Gamma target value along with a reference white measurement so L or Y target values could be calculated for each step in the grayscale. This is all done in the AccuPel Calculator tool and in CalMAN without ever coming out and expressing the actual dE number... you do get a target value for L or Y and it's up to you to adjust Y or L to get as close as possible to the target value. So you use the dE concept without ever seeing the number or how it affects the 2-D dE number you DO see... assuming you are working with something that allows adjusting L or Y for each grayscale step and primary/secondary color, like a Lumagen Radiance XD video processor. What's interesting is that there are more products out there now with 3-D control over primaries (and 2-D control over secondaries) than there are products with 10-step grayscale luminance controls. Most often we get a Gamma setting that may or may not produce the desired end result for step-by-step luminance.
dE can be 2-D or 3-D and it "works" for gray or colors - any point within a color space. It is merely an expression of how "far" a measured point is from a reference/target point within any given color space. If the reference point is expressed in 2 dimensions (xy or uv most commonly), dE is only expressing that 2-D difference and L or Y could be way off. If all 3 dimensions are included in the measurement and reference point, dE can be a 3-D expression. A dEuv formula implies that this is only a 2-D calculation, which is not influenced by L or X. On the other hand a dEuvL calculation would require all 3 coordinates (for the target and measurement) and WOULD be influenced by L or X.
We (calibrators) don't typically USE dE with grayscale measurements, but we could if the software being used was setup that way. Before starting, you'd have to enter a Gamma target value along with a reference white measurement so L or Y target values could be calculated for each step in the grayscale. This is all done in the AccuPel Calculator tool and in CalMAN without ever coming out and expressing the actual dE number... you do get a target value for L or Y and it's up to you to adjust Y or L to get as close as possible to the target value. So you use the dE concept without ever seeing the number or how it affects the 2-D dE number you DO see... assuming you are working with something that allows adjusting L or Y for each grayscale step and primary/secondary color, like a Lumagen Radiance XD video processor. What's interesting is that there are more products out there now with 3-D control over primaries (and 2-D control over secondaries) than there are products with 10-step grayscale luminance controls. Most often we get a Gamma setting that may or may not produce the desired end result for step-by-step luminance.
Joined
·
8,008 Posts
Doug:
I can't quite tell whether I agree with this or not.
First, you correctly point out that dE can be 2-D or 3-D. Using that terminology, then the correct methodology is to use 2-D measurements with the grayscale and 3-D measurements with pri./sec. colors. There are very good reasons for this which we can discuss further if you like.
Second, having made that distinction, I am not sure what you mean when you write that "We (calibrators) don't typically USE dE with grayscale measurements. . . ." I assume what you meant to say here was we don't typically use 3-D measurements for the grayscale. Again, there are very good reasons for this.
Third, I really don't know what you mean when you write that "Before starting, you'd have to enter a Gamma target value along with a reference white measurement so L or Y target values could be calculated for each step in the grayscale. This is all done in the AccuPel Calculator tool. . . ."
This is NOT how Greg's Display Calibration app handles dE for gray scale measurements. The app has two tabs, one for gray scale and one for color gamut. In the grayscale section, the dE values calculated are strictly 2-D. You can change the target gamma to anything you like and that does not affect the dE calculation, as shown below. In fact, the app allows you the ability to enter the measured Y values in the grayscale section only for the purpose of seeing the gamma. It has no effect on dE. On the other hand, under the color gamut tab, changes in L certainly do affect the calculated dE value because here the app uses a 3-D calculation. This is the correct implementation, which is why I wrote that Y is relevant for color measurements, but not for grayscale.
![]()
![]()
Doug:
I can't quite tell whether I agree with this or not.
First, you correctly point out that dE can be 2-D or 3-D. Using that terminology, then the correct methodology is to use 2-D measurements with the grayscale and 3-D measurements with pri./sec. colors. There are very good reasons for this which we can discuss further if you like.
Second, having made that distinction, I am not sure what you mean when you write that "We (calibrators) don't typically USE dE with grayscale measurements. . . ." I assume what you meant to say here was we don't typically use 3-D measurements for the grayscale. Again, there are very good reasons for this.
Third, I really don't know what you mean when you write that "Before starting, you'd have to enter a Gamma target value along with a reference white measurement so L or Y target values could be calculated for each step in the grayscale. This is all done in the AccuPel Calculator tool. . . ."
This is NOT how Greg's Display Calibration app handles dE for gray scale measurements. The app has two tabs, one for gray scale and one for color gamut. In the grayscale section, the dE values calculated are strictly 2-D. You can change the target gamma to anything you like and that does not affect the dE calculation, as shown below. In fact, the app allows you the ability to enter the measured Y values in the grayscale section only for the purpose of seeing the gamma. It has no effect on dE. On the other hand, under the color gamut tab, changes in L certainly do affect the calculated dE value because here the app uses a 3-D calculation. This is the correct implementation, which is why I wrote that Y is relevant for color measurements, but not for grayscale.
Joined
·
3,587 Posts
Yes, of course, 3-D dE is not used with grayscale adjustments. Typo on my part, should have specified.
You cut the quote off a little prematurely...
"... and in CalMAN without ever coming out and expressing the actual [3-D] dE number... you do get a target value for L or Y and it's up to you to adjust Y or L to get as close as possible to the target value."
My point being, a 3-D dE COULD have been used - everything you need for the calculation is there, it's just that there was a decision made to use a 2-D dE number with L or Y target and measurement shown separately. Which is the right thing to do when there are no controls for L or Y for each grayscale step - which is either universal or nearly universal in video displays. A device like the Lumagen Radiance XD has 3-D adjustments for each step in the grayscale but a 3-D dE number doesn't tell you anything about the direction you are "off" in (nor does a 2-D dE number for that matter) so using a 3-D dE number even if the device has 3-D controls isn't going to make the cal any easier. The best tools give you graphical or data feedback so you know whether a color is dominant or luminance is off... makes it easier to know which control to adjust and what direction to go in.
My point is/was that Y is as relevant for grayscale as it is for color. Only we tend to think of Y in terms of Gamma and that historically we haven't had step-by-step control of Y. Including Y in color measurement dEs is no more or less helpful to calibrators than not including Y in the color dEs. You could just as easily have 2-D dE calculations for color and a separate Y - for a truly functional CMS with 3 sliders per color, that would actually make adjusting the color points easier for R, G, and B since each of those colors would have a slider that affects Y only (you hope).
In the end, a 3-D dEs for grayscale measurements would/will tell you exactly the same thing as a 3-D dE for a color... how far the measured point is from the reference point. For colors, your reference point is established by SMPTE-C or Rec 709. For grayscale, your xy reference points are always the same and your Y reference point is established by your white and/or black points and your target Gamma. 3-D dEs are problematic for that simple reason that if the dE is 6... the error can be in any direction. Just because they are used with colors doesn't make them any easier to use... frankly, I'd be perfectly happy with 2-D dE plus Y for colors also.
I guess my point was that dE is a simple expression that can be 2-D or 3-D in any color space. What we do with it as calibrators is not the limitation - it's a convention, either habitual or for good reason (or both). A point on the grayscale can just as easily have a 2-D or 3-D dE number associated with it if that's what you want to do - though you may have to make your own spreadsheet to do so. [in fact, you could even have a 1-dimensional dE... a dE for x, a dE for y, a dE for Y - but you still wouldn't know the direction of the error].
Yes, of course, 3-D dE is not used with grayscale adjustments. Typo on my part, should have specified.
You cut the quote off a little prematurely...
"... and in CalMAN without ever coming out and expressing the actual [3-D] dE number... you do get a target value for L or Y and it's up to you to adjust Y or L to get as close as possible to the target value."
My point being, a 3-D dE COULD have been used - everything you need for the calculation is there, it's just that there was a decision made to use a 2-D dE number with L or Y target and measurement shown separately. Which is the right thing to do when there are no controls for L or Y for each grayscale step - which is either universal or nearly universal in video displays. A device like the Lumagen Radiance XD has 3-D adjustments for each step in the grayscale but a 3-D dE number doesn't tell you anything about the direction you are "off" in (nor does a 2-D dE number for that matter) so using a 3-D dE number even if the device has 3-D controls isn't going to make the cal any easier. The best tools give you graphical or data feedback so you know whether a color is dominant or luminance is off... makes it easier to know which control to adjust and what direction to go in.
My point is/was that Y is as relevant for grayscale as it is for color. Only we tend to think of Y in terms of Gamma and that historically we haven't had step-by-step control of Y. Including Y in color measurement dEs is no more or less helpful to calibrators than not including Y in the color dEs. You could just as easily have 2-D dE calculations for color and a separate Y - for a truly functional CMS with 3 sliders per color, that would actually make adjusting the color points easier for R, G, and B since each of those colors would have a slider that affects Y only (you hope).
In the end, a 3-D dEs for grayscale measurements would/will tell you exactly the same thing as a 3-D dE for a color... how far the measured point is from the reference point. For colors, your reference point is established by SMPTE-C or Rec 709. For grayscale, your xy reference points are always the same and your Y reference point is established by your white and/or black points and your target Gamma. 3-D dEs are problematic for that simple reason that if the dE is 6... the error can be in any direction. Just because they are used with colors doesn't make them any easier to use... frankly, I'd be perfectly happy with 2-D dE plus Y for colors also.
I guess my point was that dE is a simple expression that can be 2-D or 3-D in any color space. What we do with it as calibrators is not the limitation - it's a convention, either habitual or for good reason (or both). A point on the grayscale can just as easily have a 2-D or 3-D dE number associated with it if that's what you want to do - though you may have to make your own spreadsheet to do so. [in fact, you could even have a 1-dimensional dE... a dE for x, a dE for y, a dE for Y - but you still wouldn't know the direction of the error].
Joined
·
8,008 Posts
Doug: I cut the quote off at that point intentionally. I don't dispute that CalMan offers this as an option. I have already argued at length with Bill Blackwell about this. What I disputed was your misleading claim that CalMan AND Greg's Display Calibration app offer this as though there were some general consensus on the point, when in fact CalMan is an outlier on this issue. I merely objected to your lumping them together in this regard as though they were the same. OK, enough on that.
I said in my last post that I wasn't sure if I disagreed with you or not. Now I am sure I do. You make several claims here.
1. You COULD use 3-D dE for grayscale if you wanted to to.
2. The only reason that 3-D isn't used for grayscale is because of a lack of controls.
3. Y is equally important for grayscale dE as it is for color dE.
I think that these are fair summaries. If not, then please correct them. Let's take them one at a time.
1. Of course, one can use 3-D dE for grayscale. The fact that CalMan does it is conclusive evidence that it is possible. This issue isn't whether it is technically feasible, but whether doing so is a good idea. I think that it is abundantly clear that it is not. So clear, in fact, that I am surprised that I find myself having this debate for the second time in a month.
2. The lack of controls is not relevant at all. There are, in fact, MORE options for control of the Y parameter for white in high-end AV equipment than there is for color. Point-by-point gamma adjustment, such as what the Lumagen offers is available in other products as well, including the new JVC line of LCoS PJs. However, this degree of control of the brightness of color is unheard of. Many CMS's leave Brightness off entirely, and the ones that offer some adjustment (Lumagen incuded) offer it at one global level only. So, the industry has not adopted the practice of using Y in grayscale dE, despite the presence of quite a bit of adjustibility, especially relative to what's been available for color.
3. The issue isn't whether the Y component is important for grayscale in general. Of course, it is. The issue is whether the Y component should be included in dE calculations for the grayscale. These are related, but separate, issues. I only dispute the latter.
So it all comes to whatever reasons can be offered for or against doing so. I think that the reasons are overwhelmingly in favor of NOT doing so.
1. If you include Y in dE calculations for white, then it leads to a peculiar situation in which the same color of white at different levels of stimulus has a different dE, indeed radically different dEs, rendering the entire metric effectively meaningless. Let's take the quite green x0.311, y0.341. This results in
R91
G103
B93
From a calibration standpoint, this is an unacceptible grayscale error by any reasonable standard. Or, is it? SMPTE has specified 4 dE (in Lab units) as the outer limit of color error. If we include Y in our grayscale calculation of dE, then, assuming a 2.2 gamma, at 20% and 30% stimulus x0.311, y0.341 is a perfectly acceptable result, because the Lab dE at those levels using this methodology would be 2.5 and 3.3, respectively, and that assumes a perfect gamma response. This is true despite the fact that at 90% stimulus, the same color of white has a dE of 7.5.
Doug, this is a ludicrous result. I know of no reputable calibrator, including you, who would accept x0.311, y0.341 as an acceptible target for gray scale calibration, but that is precisely what this methodology would have us do at the low end--that is, if we are to take dE seriously. And that's my point. If you use this methodology, then dE for white is rendered nearly meaningless.
There are other reasons as well, but this one consideration is so compelling that all other factors are secondary.
Doug: I cut the quote off at that point intentionally. I don't dispute that CalMan offers this as an option. I have already argued at length with Bill Blackwell about this. What I disputed was your misleading claim that CalMan AND Greg's Display Calibration app offer this as though there were some general consensus on the point, when in fact CalMan is an outlier on this issue. I merely objected to your lumping them together in this regard as though they were the same. OK, enough on that.
I said in my last post that I wasn't sure if I disagreed with you or not. Now I am sure I do. You make several claims here.
1. You COULD use 3-D dE for grayscale if you wanted to to.
2. The only reason that 3-D isn't used for grayscale is because of a lack of controls.
3. Y is equally important for grayscale dE as it is for color dE.
I think that these are fair summaries. If not, then please correct them. Let's take them one at a time.
1. Of course, one can use 3-D dE for grayscale. The fact that CalMan does it is conclusive evidence that it is possible. This issue isn't whether it is technically feasible, but whether doing so is a good idea. I think that it is abundantly clear that it is not. So clear, in fact, that I am surprised that I find myself having this debate for the second time in a month.
2. The lack of controls is not relevant at all. There are, in fact, MORE options for control of the Y parameter for white in high-end AV equipment than there is for color. Point-by-point gamma adjustment, such as what the Lumagen offers is available in other products as well, including the new JVC line of LCoS PJs. However, this degree of control of the brightness of color is unheard of. Many CMS's leave Brightness off entirely, and the ones that offer some adjustment (Lumagen incuded) offer it at one global level only. So, the industry has not adopted the practice of using Y in grayscale dE, despite the presence of quite a bit of adjustibility, especially relative to what's been available for color.
3. The issue isn't whether the Y component is important for grayscale in general. Of course, it is. The issue is whether the Y component should be included in dE calculations for the grayscale. These are related, but separate, issues. I only dispute the latter.
So it all comes to whatever reasons can be offered for or against doing so. I think that the reasons are overwhelmingly in favor of NOT doing so.
1. If you include Y in dE calculations for white, then it leads to a peculiar situation in which the same color of white at different levels of stimulus has a different dE, indeed radically different dEs, rendering the entire metric effectively meaningless. Let's take the quite green x0.311, y0.341. This results in
R91
G103
B93
From a calibration standpoint, this is an unacceptible grayscale error by any reasonable standard. Or, is it? SMPTE has specified 4 dE (in Lab units) as the outer limit of color error. If we include Y in our grayscale calculation of dE, then, assuming a 2.2 gamma, at 20% and 30% stimulus x0.311, y0.341 is a perfectly acceptable result, because the Lab dE at those levels using this methodology would be 2.5 and 3.3, respectively, and that assumes a perfect gamma response. This is true despite the fact that at 90% stimulus, the same color of white has a dE of 7.5.
Doug, this is a ludicrous result. I know of no reputable calibrator, including you, who would accept x0.311, y0.341 as an acceptible target for gray scale calibration, but that is precisely what this methodology would have us do at the low end--that is, if we are to take dE seriously. And that's my point. If you use this methodology, then dE for white is rendered nearly meaningless.
There are other reasons as well, but this one consideration is so compelling that all other factors are secondary.
1 - 20 of 38 Posts
