Use of Delta E based color measurement process technology in Printing Industry

Introduction

Delta-E - the color difference

You don't have to spend too long in the color management world before you come across the term Delta-E. As with many things color, it seems simple to understand at first, yet the closer you look, the more elusive it gets.

Delta-E (dE) is a single number that represents the 'distance' between two colors.

The idea is that a dE of 1.0 is the smallest color difference the human eye can see. So any dE less than 1.0 is imperceptible (as in turn the lights off and head to the pub) and it stands to reason that any dE greater than 1.0 is noticeable (as in put the coffee on, we're going to be here a while). Unfortunately - and probably not surprisingly - it's not that simple. Some color differences greater than 1 are perfectly acceptable, maybe even unnoticeable. Also, the same dE color difference between two yellows and two blues may not look like the same difference to the eye and there are other places where it can fall down.

It's perfectly understandable that we would want to have a system to show errors. After all, we've spent the money on the instruments, shouldn't we get numbers from them? Delta-E numbers can be used for:

§ how far off is a print or proof from the original

§ how much has a device drifted

§ how effective is a particular profile for printing or proofing

§ removes subjectivity (as much as possible)

Delta – E 1976

So, a bit of history is probably in order. The L*a*b* colorspace was devised in 1976 (let's just call it Lab for short) and, at the same time delta-E 1976 (dE76) came into being. If you can imagine attaching a string to a color point in 3D Lab space, dE76 describes the sphere that is described by all the possible directions you could pull the string. If you hear people speak of just plain 'delta-E' they are probably referring to dE76. It is also known as dE-Lab and dE-ab (although I'm REALLY not fond of dE-ab as it implies that only the a* and b* color components are calculated and L* is left out)

One problem with dE76 is that Lab itself is not "perceptually uniform" as its creators had intended. So different amounts of visual color shift in different color areas of Lab might have the same dE76 number. Conversely, the same amount of color shift might result in different dE76 values. Another issue is that the eye is most sensitive to hue differences, then chroma and finally lightness and dE76 does not take this into account (since Lab does not take this into account).

Difference vs Tolerance

If difference is a number showing how 'far apart' two colors are, tolerance is the meaning of the number. Setting a tolerance level (such as 2.0 dE76) defines what you will accept and what you will reject(reproduction tolerance). The available differencing equations will also produce different shaped 'tolerance regions'.

Delta-Lab and Delta-LCH

One type of difference calculation that some people use is delta-L, delta-a, delta-b (dLab). By breaking the error into its components you can sometimes get a feel for what might be causing the error. If the tolerance region for dE76 is described as a round sphere, then dLab is a square cube. My favorite variation on this idea is delta-LCH. Remember that LCH is Lightness (the same one as in Lab), Chroma (the distance out from the neutral axis - saturation) and Hue (the angle/direction in the 360 degree range). If d-Lab is a box-shaped region then d-LCH is a wedge - like cutting a piece of a flat round ring or washer. The interesting thing about d-LCH is what it can tell you about inkjet behavior. Different LCH values can refer to different problems, for instance:?

§ larger dL may be a paper difference?,

§ larger dC may be paper coating difference?,

§ larger dH may be an ink difference.

As the eye's sensitivity to hue, chroma, and then lightness differ, the tolerance region around each color that contains acceptable color matches is best represented by an rugby ball-shaped ellipsoid. The more modern color difference formulae use this ellipsoid shape and allow you, the user, to vary several different parameters to tune the numbers to match visual comparisons.

How do you put a number to perceived difference in color?

No doubt a fun question to think about, but a more serious topic for print and textile business that require color consistency.

The International Commission on Illumination (CIE) formed in the early 1900s to standardize the fields of colorimetry, photometry, and imaging. The commission addressed the topic of color difference in 1976, introducing the world to the concept of Delta E.

In this guide we will take a look at three of the Delta E algorithms produced: dE76, dE94, and dE00.

Defining Delta E

ΔE - (Delta E, dE) The measure of change in visual perception of two given colors.

Delta E is a metric for understanding how the human eye perceives color difference. The term delta comes from mathematics, meaning change in a variable or function. The suffix E references the German word Empfindung, which broadly means sensation.

On a typical scale, the Delta E value will range from 0 to 100.

Delta E

Perception

<= 1.0 Not perceptible by human eyes.

1 - 2 Perceptible through close observation.

2 - 10 Perceptible at a glance.

11 - 49 Colors are more similar than opposite

100 Colors are exact opposite

Take the table as a general guide; it’s possible to get a Delta E value below 1.0 for two colors that appear different. This is the case with CIE76 and CIE94 formulas, in which saturation is either not considered or not weighted properly.

Because of inconsistencies between the three algorithms, the exact meaning of Delta E changes slightly depending on which formula is used. Think of Delta E less as a definitive answer, and instead a helpful metric to apply to a specific use case.

Delta E 76

If you were asked to write a formula to quantify perceived color difference, how would you do it?

One logical approach would be find a color matrix that is grouped on perceived color, and very simply measuring the distance between the two points in 3D space. Only one problem stands in your way: before 1976, this color space didn't exist!

CIE gave two gifts in 1976: the CIELAB color space, and the first Delta E formula.

Look familiar? Indeed, you're looking at a ripoff of the Euclidean Distance formula. This makes sense, as CIELAB color space was created as a 3D matrix of color points. Directly measuring distance should work, right?

Two points in LAB color space

Alas, issues are afoot. The premise that LAB color space is perfect in perceptual uniformity falls short, particularly with differences in saturation. The example above demonstrates that for hues in the same lightness it works well. But what happens when we have two saturated colors of different hues?

Two very different hues, highly saturated

A closer comparison of the two colors above. Are they similar? ..indeed - they appear to be!

The two above blocks represent the dark blue and dark red as graphed. CIE76 reports this difference as ΔE 10 (perceptible at a glance), when by definition difference should fall in the ΔE~1-3 range (perceptible with close observation).

Clearly saturation is a major problem for dE76. In the below example, blue highlights pixels that each algorithm has determined as 25% within tolerance of pure saturation (black).

dE76 performs poorly with high saturation comparisons.

Delta E can mean different things for a given use case. Here, if we wanted to come closer to matching all black color in CIE76, we would pad the Delta E number more. (Note this lends to more error, as tolerance increases.)

So when do we still use CIE76? The formula has the major advantage of being a simple Euclidean Distance calculation: it’s faster than the successive CIE formulas. It’s used in performance-intensive situations that don’t require high accuracy. Image processing and real-time post processing of media is an example of where CIE76 would suffice.

With such an obvious flaw around saturation, CIE continued in the laudable quest for accuracy.

Delta E 94

In 1994, the original Delta E formula was improved. The new formula would take into account certain weighting factors for each lightness, chroma, and hue value. It also introduced the ability to add a modifier according to the use case: either textile, or graphic arts.

dE94 Formula

CIE94 introduced a conversion of the given Lab value into CIE L*C*h (Lch). The two color models differ in that Lch represents hue as an angle instead of infinite points of color. This allows us to more easily troubleshoot and perform calculations on hue.

Groups hue on 4 color axes; Lch on angle

CIE94 surely outshines CIE76, but it’s no gold. The CIE94 still falls short when calculating the perceived lightness of two colors.

Hue is not similar, but the lightness is.

Given the two colors above, we can agree that the two are much different in hue, but similar in overall lightness. CIE94 calculates ΔE 128 (polar opposites), while CIE2000 more correctly calculates ΔE 49.4 (middle of the road).

CIE94 is very much Baby Bear in Goldilocks lore: It’s a middle of the road formula where accuracy is necessary but not mission-critical. It’s still often used in textiles and printing applications today.

Delta E 2000

The CIE organization decided to fix the lightness inaccuracies by introducing dE00. It’s currently the most complicated, yet most accurate, CIE color difference algorithm available.

dE00 Formula

All this busy work gets us a more accurate lightness calculation. We can see the difference in saturation weighting below. This graph compares dE94 and dE00 in posing the question, "Is pure green more similar to white or black?"

dE00 correctly changes as lightness changes. dE94 - not so much.

Bravo for dE00, which maintains the correct weighting as lightness fades from black to white. The formula concludes color difference between green/black and green/white is moderate.

dE94 incorrectly concludes that black/green has roughly the same color difference as white/green. It’s almost as if it couldn't weight saturation properly!

dE94 made a mistake here - wouldn't you say?

Although the most accurate we have today, dE00 is not without fault. The most significant discontinuity occurs when compared hues are 180° from each other (forcing a hue shift in the calculation). According to a Rochester CIE2000 Analysis paper, this could create a discontinuity of ΔE 0.2734. For all but the most high accuracy use cases, this is negligible.

Remember: Delta E accuracy must be confirmed through the very tool it was meant to remove subjectivity from - a pair of human eyes.

Choosing The Right Tolerance(Billmeyer 1970/1979)

1.   Select a single method of calculation and use it consistently

2.   Always specify exactly how the calculations are made

3.   Never attempt to convert between color differences calculated by different equations through the use of averaging factors

4.   Use calculated color differences only as a first approximation in setting tolerance, until they can be confirmed by visual judgements - in other words, verify all calculations visually

5.   Always remember that nobody accepts or rejects color because of numbers - it's the way it looks that counts.

References

1.    (2014). Retrieved from https://www.cie.co.at/

2.    Sharma, G. (2004). The CIEDE2000 Color-Difference Formula: Implementation Notes, Supplementary Test Data, and Mathematical Observations. Retrieved fromhttps://www.ece.rochester.edu/~gsharma/ciede2000/ciede2000noteCRNA.pdf

3.    Color Differences & Tolerances: Commercial Color Acceptability. (2013, January 1). Retrieved from https://industrial.datacolor.com/support/wp-content/uploads/2013/01/Color-Differences-Tolerances.pdf

4.    Brainard, D. (2003). Color Appearance and Color Difference Specification. In The Science of Color (pp. 192 - 213). Elsevier.