SAE J1545 RECOMMENDED PRACTICE
Instrumental Color Difference Measurement for Exterior Finishes, Textiles and Colored Trim
This standard was a break through document when it was originally issued in 1986 since it dealt with two important issues impeding instrumental color control at the time.
First, there was no commonly used or accepted method for calculating a DE that was usable over a wide range of colors.
As early as the 1930's it had been shown by the famous MacAdam Ellipses (color ovals) that visual acceptability plots about a standard take an elliptical shape.
Second, reference and working standards in the automobile industry seldom duplicated the official standards thus making them unreliable targets for numerical pass/fail analyses.
It is important to note the large size differences among the ellipses for
different areas of color space. For example, acceptability numbers for reds
would definitely not be suitable for greens. Blue tolerances would appear to
only work for blues, etc. This definitely illustrates the lack of visual uniformity
of 1931 CIE Color space. It can also be shown, although to a lesser degree,
that 1976 CIELab has similar problems. Additionally, DE calculated with this
system results in circular acceptability shape - a sphere centered about the
standard with a radius of DE.
The general orientation of the ellipses are such that the acceptance
deviation is more restrictive for changes in hue, less for chroma changes,
and least in lightness direction. Thus the true shape of acceptability is
an ellipsoidal volumetric shape similar to a partially deflated football that
is squeezed in at its sides making it taller than it is wide.
The recommendation was to establish pass fail tolerances for
DL*, DC*, and DH* with each color having its own set of numbers, e.g. DL*
= ± 0.85, DC* = ± 0.60, and DH* = ± 0.45, while tolerances
for another color might be ± 0.70, ± 0.42, and ± 0.25.
Using this method will enable accommodating both the non uniformity of color
space and preferential human color discrimination. There is a downside to
this method though since it produces acceptability shapes as rectangular boxes.
Try as you might, one can't fit ellipsoids into boxes or vice versa without
some gaps at the corners.
Depending upon the scaling of the tolerances one can have the box completely
within the ellipsoid, have the ellipsoid completely within the box, or somewhere
in between. In any case there will always be some gray areas of doubt that are
either overly restrictive or visually unacceptable. However, in either situation
we have a better fit to reality than using single number 1976 CIELab DE acceptance
criteria illustrated by the circles with red areas of doubt. Remember that we
are showing a two dimensional slice for ease of illustration and that the degree
of uncertainty is most likely worse than it looks.
Now that we have a method for using instrumental pass/fail analyses,
what do we use for reference and working standards? Quite simply, we use the
reference/working standards with pass/fail tolerances skewed as needed to
realign acceptability to official standard.
If the approval is close to
official standard then we can use symmetrical offsets. However, as is usually
the case, we miss the target and end up with the situation where a large part
of the pass judgments fail
to be visually acceptable to the official standard. The recommendation is to
calculate offsets and create asymmetrical P/F tolerances illustrated by the
orange rectangle now being in sync with the official standard. Acceptable parameters now might be DL* = ± 1.5 , DC* = + 0.7 - 1.3, and DH* = + 0.7 - 0.3. This procedure
is quite a bit of work but, once proper tolerances are discovered, works quite
well as long as there is some procedure to preclude the existence of metamerism.
Using pass/fail analysis in a single illuminant fails to address any metamerism
that might exist. Be very cautious using this, or any other single illuminant technique if pigmention and/or polymer differences exist among your batches and standards!
However, don't dismiss metamerism just because the standard and batch
combinations are identical. The common belief that metamerism is eliminated
by using the same pigments is not necessarily true in all cases. It's only valid
if one, and only one ratio of colorants can produce a match in one light. This
unique single light match can be guaranteed through formulating and using the
proper selection of colorants. White, black and two colored pigments is always
the preferred choice if available. Next would be white and at most, three other
colors. Any other combinations using white and more than three colors not only
complicates the shader's job, but will most likely result in what Ray Winey,
the old master who fostered my lifetime interest in color science, called system
metamerism. Computerized correction programs handle this situation OK, but visual
matchers have difficulty deciding which to add when more than one colorant can
produce the same result. Select the wrong one and bad things can happen. Add
a colorant that wasn't there originally and get even worse results.
System metamerism is easily demonstrated by making a brown color using white, black, red, and yellow. Then make a one light match for this color using the same white, black, and yellow, but substitute orange for the red. These two batches will show significant metamerism in other lights. Blend these two batches in ratio of 90/10 and 10/90 and you will have system metamers - conditionally variant matches containing the same pigments!
The same era that produced the original SAE J1545 also spawned a visually derived, ellipsoid producing method for calculating color differences from 1976 CIELAB data. This procedure, an offspring of JPC79, was first standardized in England as a British Standard and later adopted in America by the AATCC, came to be known as CMC after the Colour Measurement Committee of The Society of Dyers and Colourists that formalized it. Pictured below are CMC acceptability ellipses plotted on 1976 CIELAB a*b* graph showing relative size and shape changes with movement through color space.
Through the late '80's and early '90's CMC began to work its way into color
software and started to gain some acceptance as a "one number for all colors"
method. Different people working independently in varied industries all seemed
to arrive at the same conclusion - a DECMC
around 0.4 to 0.5 identified a critically acceptable visual match for a surprisingly
large number of colored pairs. Ford conducted a very early experiment for total
instrumental color control using 0.40 as the acceptance criteria. This program
produced millions of molded parts without a single off color incident adding
much credence to use of CMC. Other programs outside the auto industry were equally
successful. In the decade+ since these experiments, use of CMC has become fairly
commonplace by those who have kept up with the technology. The evolution of
better instrumentation, faster and cheaper computers, more widespread availability
of color systems, and recent developments for dealing with gonioapparent materials
all indicate that the original SAE J1545 needed to be brought into the 21st
This will not be a simple task. Instead of endorsing CMC, CIE issued their
own improvement to CIELAB known as CIE94 derived from selected color evaluation
data sets available at the time. CMC was excluded because a large part of the
original work was judged by one person and subsequent evaluations involved a
half dozen or so individuals. In order to improve the models for future color
difference evaluations, viewing conditions were set with emphasis placed upon
the lightness contrast between the samples and the background. This reportedly
caused some problems with CMC. Another questionable area is the band of narrow
ellipses contained in the brown triangle that appear to be hue angle related
but not verified by the CIE94 work. The additional reported CMC weakness needing
improvement is the orientation of the ellipses in the blue rectangle. Data indicates
they should be rotated somewhat. The CIE technical committee charged with improving
or replacing CIE94 completed their task with the issue of CIE DE2000. To further
muddy the waters, a separate group, also seeking the ultimate color difference
system issued DIN99. It seems we should have a "one number works for all colors"
way of calculating DE. The question is, which one? CMC,
DE2K, DIN, and CIELAB compared here.
We now have a new problem. Since DE calculated by any method is a scalar quantity, not having plus or minus directions, we cannot skew tolerances to account for missing the target with our reference standard. One choice is to measure the official standard. However, there are many instances when we should not measure our batches vs. the OS. (See link to more detailed information at bottom of page.) It is possible to calculate offsets that would be applied to the measurements prior to calculating DE but I know of no QC software that makes this provision. Unless this is done for multiple illuminants, we are still subject to the previous limitation for detecting metamerism. A much better solution would be to shift the reference standard to coincide with the OS. This is done by a technique that we call curve shifting.
Minor adjustments are made to the curve without appreciably changing its shape. This allows moving its calculated LAB coordinates without creating any significant metamerism that would make it useless as a reference standard. Curve synthesis, a capability often found in computerized formulation programs, involves manipulating colorant K and S data to obtain the desired reflectance numbers. It is helpful to have some knowledge of the colorant content of your standards when using this process. This method was used in the Ford program mentioned earlier. Direct manipulation of the reflectance values to produce the shifted curve is also possible. This technique requires no knowledge of pigmentation or K and S data. The shifted curve shown here was produced by this technique.
Having a curve as a standard has overwhelming advantages over any other methods
for handling these situations. A curve enables batches to be evaluated in any
desired illuminant or combination of illuminants thus detecting all forms of
metamerism. It goes without saying that comparison of curves is the best indicator
of formula integrity. A curve enables the effective use of computerized color
correction and formulation systems. Curve data is compatible with all existing
programs. Curves are really the basis of all colorimetry. Color doesn't exist
without a curve!
your questions, suggestions or other comments.