Understanding dE - AVS Forum
Forum Jump: 
 
Thread Tools
post #1 of 12 Old 09-30-2005, 01:32 PM - Thread Starter
Scott Horton, techht.com
 
GetGray's Avatar
 
Join Date: Aug 2004
Location: Mid-South USA
Posts: 5,451
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 30 Post(s)
Liked: 52
I was hoping some of the regular more experienced guys might chime in with some explanation and discussion about dE. How it is measured/expressed is a little fuzzy to me. ?

Thanks guys.

Scott
GetGray is offline  
Sponsored Links
Advertisement
 
post #2 of 12 Old 09-30-2005, 01:49 PM
umr
AVS Club Gold
 
umr's Avatar
 
Join Date: Oct 2002
Location: USA
Posts: 10,138
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 4 Post(s)
Liked: 29
Delta E is a color/luminance tolerance metric. The luminance portion is normally ignored when used in this arena. There are various methods available for calculating it that have evolved with time.

Here are some links that show the different calculation methods.

http://www.brucelindbloom.com/index....eltaE_CMC.html

http://www.brucelindbloom.com/index....taE_CIE94.html

http://www.brucelindbloom.com/index....taE_CIE76.html

A zero Delta E difference should not be distinguishable as different from another sample. A difference of one Delta E should be bairly distinguishable from the other sample as being a different color/intensity. The goal is also to have the metric indicative of the difference as perceived by a person. Meaning that a 10 should be a worse match than a 5. This is of course the ideal situation and not always the case.
umr is offline  
post #3 of 12 Old 09-30-2005, 02:41 PM
AVS Special Member
 
Ursa's Avatar
 
Join Date: Jul 2003
Location: Inner Loop, Houston
Posts: 5,001
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Jeff has it. Poynton just uses the 1976 version, and as Bruce Lindbloom points out, it is flawed.

The 1976 method is heavily dominated by the L* component of either the CIELab or CIELuv coordinate systems. Thus, grayscale error is represented in both the dL* component, and in the du* and dv* (or da*, db*) calculations. "Delta C" (dC*) eliminates the L* component, but does not eliminate the impact of luminance entirely from the calculation (L* is used as a scaler when converting xy -> u'v' -> u*v*).

Later,
Bill
Ursa is offline  
post #4 of 12 Old 09-30-2005, 02:42 PM
AVS Addicted Member
 
krasmuzik's Avatar
 
Join Date: Oct 1999
Location: NewPort, VA
Posts: 11,270
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 11
I do not agree with using CIELAB for displays - as it was more intended for the paint and textile industry where the apparent surface color varies depending on the color temperature of the illumination. CIELUV was intended for the light source itself. This is why CIE 1976 could not settle their differences - and they picked both color specification. And most of the improved color formulaes have worked on the surface color issues.

An excellent resource for those that want to play with color formulas is

http://www.wiley.com/WileyCDA/WileyT...470845627.html
krasmuzik is offline  
post #5 of 12 Old 09-30-2005, 03:18 PM
umr
AVS Club Gold
 
umr's Avatar
 
Join Date: Oct 2002
Location: USA
Posts: 10,138
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 4 Post(s)
Liked: 29
Quote:
Originally Posted by krasmuzik
I do not agree with using CIELAB for displays - as it was more intended for the paint and textile industry where the apparent surface color varies depending on the color temperature of the illumination. CIELUV was intended for the light source itself. This is why CIE 1976 could not settle their differences - and they picked both color specification. And most of the improved color formulaes have worked on the surface color issues.

An excellent resource for those that want to play with color formulas is

http://www.wiley.com/WileyCDA/WileyT...470845627.html
Kevin,

I think you will find the differences pretty subtle in most cases. Here is an example of eleven different measurements with calculated Delta E values for CIELUV values scaled to approximate CIE 1976. These are just color tolerancing functions with a goal of zero in every case. It appears to me the method selected is not a critical choice in the calibration process. I do believe using one is a good idea though.


Delta E u' v'

37
21
11
8
7
7
6
5
6
6
6

Delta E 1976

42
21
10
8
8
7
6
5
6
5
6

Delta E 1994

15
11
7
6
6
5
5
4
4
4
5

Delta E CMC

18
12
8
7
7
7
6
5
6
6
6
umr is offline  
post #6 of 12 Old 09-30-2005, 04:03 PM
AVS Special Member
 
gregr's Avatar
 
Join Date: Nov 2001
Posts: 2,128
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Quote:
Originally Posted by Ursa
Jeff has it. Poynton just uses the 1976 version, and as Bruce Lindbloom points out, it is flawed.

The 1976 method is heavily dominated by the L* component of either the CIELab or CIELuv coordinate systems. Thus, grayscale error is represented in both the dL* component, and in the du* and dv* (or da*, db*) calculations. "Delta C" (dC*) eliminates the L* component, but does not eliminate the impact of luminance entirely from the calculation (L* is used as a scaler when converting xy -> u'v' -> u*v*).

Bill
When we use dE* as a metric for grayscale tracking we wish to describe the color difference between an actual grayscale color and the D65 reference color, both at the same lightness. Therefore L* = 100 and delta L*=0. Hence, L* becomes a fixed scale factor in the dE* calculation. If we want to express luma errors in the grayscale tracking we show that separately as a gamma error.

Greg Rogers
AccuPel
Widescreen Review

Greg Rogers
Video Engineer/Product Designer

gregr is online now  
post #7 of 12 Old 09-30-2005, 04:11 PM
umr
AVS Club Gold
 
umr's Avatar
 
Join Date: Oct 2002
Location: USA
Posts: 10,138
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 4 Post(s)
Liked: 29
Quote:
Originally Posted by gregr
When we use dE* as a metric for grayscale tracking we wish to describe the color difference between an actual grayscale color and the D65 reference color, both at the same lightness. Therefore L* = 100 and delta L*=0. Hence, L* becomes a fixed scale factor in the dE* calculation. If we want to express luma errors in the grayscale tracking we show that separately as a gamma error.

Greg Rogers
AccuPel
Widescreen Review
That is exactly the way I use it as well.
umr is offline  
post #8 of 12 Old 09-30-2005, 04:35 PM
AVS Special Member
 
Ursa's Avatar
 
Join Date: Jul 2003
Location: Inner Loop, Houston
Posts: 5,001
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Depending upon which calculation method you use, eliminating dL* is the same as dC*. In a CIELuv coordinate system, setting L* to 100 means that the dE* you are calculating is the maximum perceivable error, and not the actual error (e.g., l* for a "pure" green is ~87, so the error in a pure green may be overstated by ~15% just on calculation issues alone). Personally, I would argue that if you want dC*, use dC*.

We can talk about gamma computation in another thread...

Later,
Bill
Ursa is offline  
post #9 of 12 Old 09-30-2005, 08:57 PM
AVS Addicted Member
 
krasmuzik's Avatar
 
Join Date: Oct 1999
Location: NewPort, VA
Posts: 11,270
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 11
Ursa

Yes I think the reference brightness of the target you are matching should be used - be it Green, Black or White or Grey - they all have different L* values. Obviously Black has zero perceivable error - because it is black! So that brings into maybe using the measured L* to scale the error - so that DLP black is the dark grey it is - in which color error is more perceivable.

umr
For the calibration process - it may not matter because the dE used may not matter as it gets you dialed in. But then this gets similar to the RGB% discussion we had before - it is just a tolerance measure. I am more worried about using something consistent with professional video engineers - so that when a review says something is 20dE off out of the box - everyone knows what that means.

I need to look it up - but I recall reading somewhere that it said some of the dE measures for surface color were to provide greater accuracy when the other measures read <3dE - even though they get wacked out when the accuracy is way off. But for paint matching they wanted something more precise.
krasmuzik is offline  
post #10 of 12 Old 10-02-2005, 10:13 AM
AVS Addicted Member
 
TheFerret's Avatar
 
Join Date: May 2004
Location: The Swamps of Georgia
Posts: 10,004
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Hey Scott, was there a context (i.e. r u renting colorfacts or something) to apply your question to?

KRAS = Kevin?

Jeff, you quoted Greg Rogers but did he remove his post, or was that from another thread?

When will D* stop pushing HD-Lite while charging us for full HD? Digital input on a CRT is a reality, not a possibility.
TheFerret is offline  
post #11 of 12 Old 10-02-2005, 11:26 AM
AVS Special Member
 
Ursa's Avatar
 
Join Date: Jul 2003
Location: Inner Loop, Houston
Posts: 5,001
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 0 Post(s)
Liked: 10
Ursa is offline  
post #12 of 12 Old 10-02-2005, 01:47 PM
umr
AVS Club Gold
 
umr's Avatar
 
Join Date: Oct 2002
Location: USA
Posts: 10,138
Mentioned: 0 Post(s)
Tagged: 0 Thread(s)
Quoted: 4 Post(s)
Liked: 29
Quote:
Originally Posted by TheFerret
...Jeff, you quoted Greg Rogers but did he remove his post, or was that from another thread?
That was from this thread.
umr is offline  
Closed Thread Display Calibration

User Tag List

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


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