Delta E or by the numbers. what to use. - Page 2 - AVS Forum
Forum Jump: 
 
Thread Tools
Old 01-09-2009, 11:50 AM
AVS Club Gold
 
Bear5k's Avatar
 
Join Date: Feb 2006
Location: Houston, TX
Posts: 2,239
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by TomHuffman View Post

Bill, CIE establishes general standards for color. It does not get into the weeds of how specific industries implement those standards. If the CIE has published standards for how to use dE in grayscale evaluation, then I'd love to read it.

I also reference SMPTE, which you have ignored. Please read SMPTE Engineering Guideline 432-1, especially appendix L. I will quote the relevant sections here to save folks the cost of buying it from the SMPTE store. I HIGHLY recommend that pro calibrators join SMPTE, if only for the discounts for SMPTE publications.

So that I do not create a monster post (it's going to be big, as it is), and I stay within fair use doctrine for copyrighted materials, I am only going to post excerpts. If you want to double-check that I am not quoting things out of context, please purchase the document I reference for yourself. Note: you can get some of this content, but with some odd contradictions, from www.dcimovies.com, though do realize that SMPTE Engineering Guidelines are authoritative.

Quote:


If you are correct, then the entire video industry ignores CIE. Not just me. Again, this is not personal. I don't have a beef against you or against Derek. Aside from being more than a little over-sensitive, you seem like a nice guy. This a substantive disagreement not a pissing match.

So, let's try to establish a few things:
1) What are the properties of a color error equation?
2) Should there be any differences as it applies to grayscale and why?
3) What is the best way to quantify these adjustments?

Let's turn to Appendix L, it's a good treatment of a topic I have, so far, been unsuccessful in getting you to understand:

Quote:
Originally Posted by SMPTE EG-432-1-2007, pp. 72 - 73 View Post

The Munsell color space is the most uniform color space. - It was defined by having many people make many judgements of color differences using color patches. One problem with the Munsell space is that it was based on judgements of color patches and it is not based on any mathematical equations. The CIELab space is a mathematical approximation, based on XYZ tristimulus values and Equations L-1 to L-13, of the Munsell space. But the equations that define the CIELab space do not exactly describe the Munsell space. Therefore the CIELab space is not perfectly uniform, but is sufficiently uniform for most uses.


So, one of the issues we are dealing with is that we have to give up some perceptual uniformity for quantification and ease-of-modeling. The implications here is that at some places, the perceptual side will break down, but the benefits gained in terms of predictive capabilities outweigh these point failures. In other words, nothing is perfect.

Further in the same paragraph:
Quote:


There is a general rule of thumb that says that when comparing two color patches, which are placed near each other, which are near neutral in color, and which are in the environment specified for judging colors in the Munsell color system, a delta E*ab = 1 is at or near the threshold of visibility of the color difference for most people. If any of these conditions is changed, for example the environment is changed or the colors are very colorful instead of being near neutral or the colors are presented sequentially at very low frequency, not simultaneously, the threshold of visibility of the color difference increases. This means that for any of these different conditions, in order for a person to see the color difference, the delta E*ab will increase. Conversely, for any of these different conditions, a pair of color patches with a given delta E*ab will appear to be less different than if they were in the environment specified for judging colors in the Munsell color system. For the case in which the two colors are highly colored, the threshold of the visibility of the color difference increases to a delta E*ab of about 2. The result of this is that because the illuminance levels in a theatre are much lower than were used to define the Munsell space and because the color differences are between colors in one theatre using one projector and colors in another theatre using another projector, this delta E*ab of 4 is visually a very tight tolerance. In fact, few, if any, people will be able to detect the color difference between two colors from the same code values in two different theatres if both theatres and projectors are set up to these tolerances.



(Emphasis mine) While a reference back to the origin of the "4 dE" standard is not made in the text, its validity is based upon studies done by SMPTE. Do note, though, that this is for fully-specified Lab. Also note the reference to the "dim" environment in a typical theater. Color differences have been demonstrated to be harder to see where light levels are lower. Let's see how this translates back to the xy plane:

Quote:
Originally Posted by p.73 View Post

In the a*b* diagram for each primary there are four circles for the four levels of luminance around each primary. The four circles for each primary are not all centered at the same a*b* values, but instead fall on a line from the a*b* values of the primary at its highest luminance to the neutral axis. The circle corresponding to the 0.5% luminance patch is the closest to the neutral axis. In the chromaticity diagram, the reference point of the tolerance figures is located at the same xy chromaticity coordinates independent of the luminance value. However, although all of the tolerance figures were circles of radius 4 in the a*b* diagram, the tolerance figures in the xy diagram increase in size with decreasing luminance.

(Again, emphasis mine) Here we see the difference between what one gets at the periphery versus at the "neutral" (i.e., gray/white) point. Again, for copyright purposes, I am attaching only the part figure L-2 where a constant dE of 4 (with dL = 0) is re-cast from the ab plane to an xy plane. What is notable is the absence of a discussion here of using a different calculation approach to dE for work in the gamut versus the grayscale.

Having worked through the introductory material in the appendix, let's look to section 6, Measurement of Projected Images (seems relevant). On page 13, you see the chromaticity and luminance targets and tolerances for reference theaters, review rooms and general exhibition theaters. Here we get a bit of irony, there is a +/- 0.002 deviation allowed for deviation in x or y, even though the perceptibility of these deviations is not constant. In practice, these should be very small, though. These are well within the measurement confidence interval for ordinary colorimeters. For luminance, the reference is 14ftL (unambigous) with a tolerance of +/- 1 ftL for review rooms (also unambiguous).

The tolerance for luminance is given in terms of a variance on gamma, rather than a variance in dE. SMPTE uses an ordinary power curve, with the computation being done in a log-log fashion. Reference slope (i.e., gamma exponent) is 2.6, with a tolerance for reference rooms of 2.548 to 2.652 (+/- 0.52). So, this is pretty important. What have we learned:

  1. SMPTE uses dE(76, Lab)
  2. The amount of saturation/hue error that is perceptible varies with changes in adapted light level.
  3. Error in luminance is a "measure of concern" to SMPTE, but is measured in terms of gamma deviation, rather than dE -- until we get to section 6.13.
Section 6.13 gives us the method for verifying the color accuracy of the entire system using 12 measures, two each for RGBCMY. Here, the standard of dE(ab) being <= 4 is reasserted. Since white/gray is a color, presumably these also fall within this standard, though in reality, the tolerances given previously are a tighter standard.

Quote:
Originally Posted by 6.13; page 18 View Post

Within the minimum color gamut specified for the Reference Projector, all colors need to be accurately reproduced within a tolerance of 4 delta E*ab. A discussion of delta E*ab is given in Appendix L. In theory this applies to all colors, but in practice it would be impossible to display and measure all possible colors that can be encoded by the DCDM color encoding equations and displayed by a Reference Projector. Therefore Table 6-11 gives a set of colors that can be used to verify the color accuracy of a system. It is felt that if these colors are within the tolerance limits, then all colors are most likely within the tolerances. The neutral colors 6 through 10 in Table 6-7 may also be used as tests of the color accuracy of any projector in its environment.


So, given this, the obvious question is "why does CalMAN advocate anything different from SMPTE?" The reason is that until relatively recently, there were not ways for people to manipulate the gamma performance of their displays, so we wanted a method that preserves the (desirable) characteristic of color error diminishing as light level decreases, while, at the same time, giving our predominantly enthusiast customers clear guidance that was usable in a typical home environment with home tools (e.g., a 0.002 tolerance does you no good if your meter's limit under best circumstances is 0.004). In this respect, we use uv, rather than ab, because uv is better than xy, and at least a few publications (e.g., WSR) use it. Once you move beyond dE(76), you are "stuck" with Lab, so there is no uv vs. ab controversy there, only threads like this.


Quote:
Originally Posted by TomHuffman View Post

You must agree with me because that's what your software displays.

CalMAN displays a lot of things. Our goal is to provide the user as much information as possible. In terms of gamut charts, we prefer/recommend people use a uv chart, but there is only so much that education and outreach can do, as my efforts here prove.

I've said it before, and I'll say it again: dE(uv) provides a much more aggressive standard than either of the dE(76) methods. People who use it should feel confident that they are not going to encounter visible errors resulting from problems in saturation or hue. For an enthusiast, this may be too aggressive, at which point, CalMAN offers several alternatives.

If this is sufficient, then I'm not really sure what is.

Bill
LL

Color accuracy evangelist and CalMAN insider
Bear5k is offline  
Sponsored Links
Advertisement
 
Old 01-09-2009, 01:37 PM
AVS Special Member
 
TomHuffman's Avatar
 
Join Date: Jan 2003
Location: Springfield, MO
Posts: 6,479
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 80 Post(s)
Liked: 249
Quote:
Originally Posted by Bear5k View Post

If this is sufficient, then I'm not really sure what is.

Nothing you have quoted here speaks to what we have been arguing about, which is two approaches to employing dE for grayscale measurements. Indeed, nothing in this lengthy series of quotes even mentions grayscale tracking.

I think that we should just agree to disagree and leave it at that.

Tom Huffman
ChromaPure Software/AccuPel Video Signal Generators
ISF/THX Calibrations
Springfield, MO

TomHuffman is offline  
Old 01-11-2009, 01:59 AM
AVS Special Member
 
Doug Blackburn's Avatar
 
Join Date: May 2008
Location: San Francisco - East Bay area
Posts: 3,457
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 25 Post(s)
Liked: 228
Quote:
Originally Posted by cid67 View Post

It seams to me that what is going on in my case is, because my blue is desaturated I should push my Y value for blue to compensate for this and thus giving more of a perception of being closer to accurate. that's at least what delta e is telling me so that's what I have done. I can't get delta e for my blue below 5.6. all other colors are 2 or below. that's just the limit of my sets capability, I guess.

Y doesn't have anything to do with saturation though. Y is brightness. So you are undersaturated (this means, literally, that your measured blue point is INSIDE the reference color space triangle you are using... HDTV/709 or SD/SMPTE-C). All Y will do is make that shade of blue lighter or darker, it won't move the measured point farther out towards the correct reference point. So altering the Y will INCREASE your error for that point, not decrease it. Your lowest error will be with Y as close to the reference Y for Blue as you can get, and the x & y points as close to the reference x & y as you can get. If your blue really is desaturated, there's really nothing you can do to pump it up to look right. In otherwords, if you increase Y trying to compensate for the desaturated condition, you will make every color that contains blue even less accurate than they are with Y set close to the reference value for Blue Y.

"Movies is magic..." Van Dyke Parks
THX -- ISF -- HAA
Doug Blackburn is offline  
Old 01-11-2009, 09:52 AM - Thread Starter
Advanced Member
 
cid67's Avatar
 
Join Date: Apr 2007
Posts: 682
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 12
thanks for being clear. yes I am perfectly aware that Y is luminance and willl not directly effect saturation and hue. I have tried putting my blue luminance at the correct setting according to this formula

Rec. 709 (High Definition)
--------------x-------y-------Y
Red----------0.640---0.330---0.2127
Green--------0.300---0.600---0.7151
Blue----------0.150---0.060---0.0722
Yellow--------0.419---0.505---0.9278
Cyan---------0.225---0.329---0.7873
Magenta------0.321---0.154---0.2849
White--------0.3127--0.329---1.0

so basically it was done this way
100% white luminance X 0.08 = luminance for blue
this gives me a worse delta E than where I have it .

the next thing you might assume is that my meter is off but I have confirmed with acupel calculator and got close to the same result.

based on this I can only assume that even though luminance doesnt effect saturation, having a slightly brighter less saturated color is beter than having a accurate luminance with a les saturated color.

this will not through off all my other colors because I am working with a five stage (possible 6 stage) color wheel and each of the six colors can be adjusted independently and without effecting the other.

here once again I go back to my original question of weather or not to believe in delta e or go by the numbers.
cid67 is offline  
Old 01-11-2009, 11:32 AM
AVS Special Member
 
BeachComber's Avatar
 
Join Date: Jan 2008
Location: Wandering a Beach Somewhere....
Posts: 2,618
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 11
Quote:
Originally Posted by cid67 View Post

thanks for being clear. yes I am perfectly aware that Y is luminance and willl not directly effect saturation and hue. I have tried putting my blue luminance at the correct setting according to this formula

Rec. 709 (High Definition)
--------------x-------y-------Y
Red----------0.640---0.330---0.2127
Green--------0.300---0.600---0.7151
Blue----------0.150---0.060---0.0722
Yellow--------0.419---0.505---0.9278
Cyan---------0.225---0.329---0.7873
Magenta------0.321---0.154---0.2849
White--------0.3127--0.329---1.0

so basically it was done this way
100% white luminance X 0.08 = luminance for blue
this gives me a worse delta E than where I have it .

the next thing you might assume is that my meter is off but I have confirmed with acupel calculator and got close to the same result.

based on this I can only assume that even though luminance doesnt effect saturation, having a slightly brighter less saturated color is beter than having a accurate luminance with a les saturated color.

this will not through off all my other colors because I am working with a five stage (possible 6 stage) color wheel and each of the six colors can be adjusted independently and without effecting the other.

here once again I go back to my original question of weather or not to believe in delta e or go by the numbers.

The THX classes, the dE 3.0 or less (not go by the numbers as Joel teaches) even states that after everything you put up known material and the final check is a visual check using known material using your eyes.

If you say it looks wrong using dE, you have made the decision yourself.
BeachComber is offline  
Old 01-12-2009, 12:04 PM
Senior Member
 
cvearl's Avatar
 
Join Date: Sep 2008
Posts: 440
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by cid67 View Post

thanks for being clear. yes I am perfectly aware that Y is luminance and willl not directly effect saturation and hue. I have tried putting my blue luminance at the correct setting according to this formula

Rec. 709 (High Definition)
--------------x-------y-------Y
Red----------0.640---0.330---0.2127
Green--------0.300---0.600---0.7151
Blue----------0.150---0.060---0.0722
Yellow--------0.419---0.505---0.9278
Cyan---------0.225---0.329---0.7873
Magenta------0.321---0.154---0.2849
White--------0.3127--0.329---1.0

so basically it was done this way
100% white luminance X 0.08 = luminance for blue
this gives me a worse delta E than where I have it .

the next thing you might assume is that my meter is off but I have confirmed with acupel calculator and got close to the same result.

based on this I can only assume that even though luminance doesnt effect saturation, having a slightly brighter less saturated color is beter than having a accurate luminance with a les saturated color.

this will not through off all my other colors because I am working with a five stage (possible 6 stage) color wheel and each of the six colors can be adjusted independently and without effecting the other.

here once again I go back to my original question of weather or not to believe in delta e or go by the numbers.

Do the Greyramp patterns look completely nuetral before you even get into this attempt at color decoding correction and saturation/hue adjustments?

Logic dictates that if there is tinting in greyramps to your eye regardless of dE for greyscale, then everything else is crap. Greyscale has to be visually confirmed before moving on is what my mind tell me.

I have had times playing with my cuts and drives and had RGB fairly well balanced and dE's 4 or less across the greyscale run and then the temp graph showed that I was giong below D65 and above D65 from one step to the next. The ones that were D65 or above looke great in the Greysteps but the ones that dipped below D65 appeared pink and a few of them a touch green.

I am assuming that that would effect how color ends up looking right?

C.

Panasonic 42" Plasma TH42-PZ85U
Panasonic Blueray Player DMP-BD35
Sony 5.1 Home Theater DAV-HDX475 (wish it was Panasonic but it's ok)
cvearl is offline  
Old 01-12-2009, 09:59 PM
AVS Club Gold
 
Bear5k's Avatar
 
Join Date: Feb 2006
Location: Houston, TX
Posts: 2,239
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by cid67 View Post

thanks for being clear. yes I am perfectly aware that Y is luminance and willl not directly effect saturation and hue. I have tried putting my blue luminance at the correct setting according to this formula

Rec. 709 (High Definition)
--------------x-------y-------Y
Red----------0.640---0.330---0.2127
Green--------0.300---0.600---0.7151
Blue----------0.150---0.060---0.0722
Yellow--------0.419---0.505---0.9278
Cyan---------0.225---0.329---0.7873
Magenta------0.321---0.154---0.2849
White--------0.3127--0.329---1.0

so basically it was done this way
100% white luminance X 0.08 = luminance for blue
this gives me a worse delta E than where I have it .

the next thing you might assume is that my meter is off but I have confirmed with acupel calculator and got close to the same result.

based on this I can only assume that even though luminance doesnt effect saturation, having a slightly brighter less saturated color is beter than having a accurate luminance with a les saturated color.

this will not through off all my other colors because I am working with a five stage (possible 6 stage) color wheel and each of the six colors can be adjusted independently and without effecting the other.

here once again I go back to my original question of weather or not to believe in delta e or go by the numbers.

If you are trying to measure the gamut with a consumer-oriented colorimeter, go with your eyes. If you've got an i1 Pro, re-calculate the color balance ratios, and then re-take the measurements to see what your actual targets are (per Tom's suggestion from post #2).

Color accuracy evangelist and CalMAN insider
Bear5k is offline  
Old 01-12-2009, 10:28 PM
AVS Special Member
 
Doug Blackburn's Avatar
 
Join Date: May 2008
Location: San Francisco - East Bay area
Posts: 3,457
Mentioned: 1 Post(s)
Tagged: 0 Thread(s)
Quoted: 25 Post(s)
Liked: 228
Quote:
Originally Posted by cid67 View Post

thanks for being clear. yes I am perfectly aware that Y is luminance and willl not directly effect saturation and hue. I have tried putting my blue luminance at the correct setting according to this formula

Rec. 709 (High Definition)
--------------x-------y-------Y
Red----------0.640---0.330---0.2127
Green--------0.300---0.600---0.7151
Blue----------0.150---0.060---0.0722
Yellow--------0.419---0.505---0.9278
Cyan---------0.225---0.329---0.7873
Magenta------0.321---0.154---0.2849
White--------0.3127--0.329---1.0

so basically it was done this way
100% white luminance X 0.08 = luminance for blue
this gives me a worse delta E than where I have it .

the next thing you might assume is that my meter is off but I have confirmed with acupel calculator and got close to the same result.

based on this I can only assume that even though luminance doesnt effect saturation, having a slightly brighter less saturated color is beter than having a accurate luminance with a les saturated color.

this will not through off all my other colors because I am working with a five stage (possible 6 stage) color wheel and each of the six colors can be adjusted independently and without effecting the other.

here once again I go back to my original question of weather or not to believe in delta e or go by the numbers.

There are different dEs and depending on which one you are using, the "limits" you want to use are also different. dExy (there are more than 1 of these based on the CIE standard selected, usually going by the year the standard was adopted) and dEuv produce different dE numbers for the same coordinates.

Rounding .0722 to .08 will increase your error. If you are going to round .0722, going DOWN to .07 times the 100% white luminance will decrease your error.

If your meter isn't accurate, the Accupel calculator won't improve your results or confirm anything about the accuracy of the meter. Using 2 different calculation methods with a meter that's not providing accurate readings won't reveal anything - except you may get similar results for the 2 calculation methods. That doesn't mean the meter is accurate. Doesn't mean it's inaccurate either.

Have you tried using a "cooler" color temp setting to see if your blue point, just by luck, is more saturated? If the color temp setting moves the blue point in the right direction, you'll probably have better luck bringing green and red "in" towards the correct reference points.

"Movies is magic..." Van Dyke Parks
THX -- ISF -- HAA
Doug Blackburn is offline  
 
Thread Tools


Forum Jump: 

Posting Rules  
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off