Thanks for sticking with us—this is the sixth and final post about the NCCA Color Project experiment we conducted at METALCON. In the previous five posts, we presented our analyses of the 28 observers’ ratings to see how discerning and consistent they were. We concluded that human observers see color differences differently; some see a lot of difference, some just a little. This was not unexpected. Finally, it’s time to look at the observed color differences plotted against the machine readings for color difference.
Recall that we used the following rating scale throughout the experiment, and this scale is the Y-axis of the graphs that follow.
5 = No color difference
4 = Extremely slight color difference
3 = Slight color difference
2 = Noticeable color difference
1 = Very noticeable color difference
For each color pair, it made sense to look at the average ratings given by all 28 observers. In this first graph, we compare these average visual ratings versus ΔEHunter.
A quick inspection of this graph shows that there is a general downward slope, which makes sense: As the color differences become more apparent (a lower value on the Y axis), you’d expect the ΔEHunter to get larger. This is clearly a noisy trend line, one that is hardly worth the addition of a linear trend line.
Let’s look at the same graph, but instead of plotting against ΔEHunter, we’ll use ΔE2000.
I think the data is a bit cleaner, but it is clearly not perfect. Look at the two data points in the upper right-hand side of the graph. Our observers saw little color difference in these two pairs, but there is actually quite a large ΔE2000.
In an effort to attempt to clarify the picture a little bit, let’s try this:
- Let’s assume that a rating of 3 (slight color difference) or better will be acceptable. So let’s draw a horizontal line on the graph at the rating 3 line. Anything above this horizontal line will be considered an acceptable color difference; anything below, an unacceptable color difference.
- And then let’s draw a vertical line such that no data points lie in the bottom left quadrant. A data point falling into this quadrant would be visually unacceptable, but would read acceptably. This kind of false positive (reads OK; looks bad) would be an intolerable condition that must be avoided at all costs, which is why we force this bottom left quadrant to contain no data points.
Such a graph, using ΔEHunter on the X-axis, would look like this:
In the upper left quadrant, these are panels that look fine and read well. No issues here! In the bottom right quadrant, these are panels that look poor and don’t read very well, either. No issues here! But in the upper right-hand quadrant, these are panels that look good, but read poorly. You can count seven data points that fall into this quadrant.
Let’s quickly switch to a similar construction, but where the X-axis uses ΔE2000.
As you can see above, there are only two data points that are clear outliers. (That third data point near the red crosshairs is so close to being in one of the other quadrants, I’m going to ignore it for now.)
These two graphs suggest that the value of considering ΔE2000 has merit. This is hardly a ringing endorsement, although it’s the best that can be offered at this time. And it falls in line with what color experts have said: that ΔE2000 is a better color difference system than ΔEHunter. Since the foundation of ΔE2000 was based on human observations in the first place, and then the math was wrapped around it, it stands to reason that it would stand up in an experiment such as the one we have done.
NCCA has conducted an excellent experiment, and there is an abundance of data, but this writer has only limited expertise in this area. My intent is to show and discuss our data with real color industry experts to get a sense of what we’ve done.
Still, there is much more to do. Our study involved only solid colors; no metallics or pearlescents were included. We cannot ignore these colors. All the panels were flat, but many of our building products are fabricated. I’ve been told that introducing shape (such as corrugation) would confuse the eye and will allow for color pairs with greater ΔE values to be seen by observers as being a closer match than the ΔE value would suggest. ACP panels are flat, but they might have a small gap between them. Any gap might also confuse the eye. This is only the beginning …
If you would like to get into more of the details, please feel free to contact me at firstname.lastname@example.org. I am more than happy to offer further explanation, and I would be grateful to hear any suggestions you have to offer.
We’ve taken just the first step on a long journey, so stay tuned for additional analysis and, very likely, another color experiment.
NCCA Technical Director