Wednesday, November 19, 2014

Measuring fluorescent inks

People email me questions fairly reguarly. If I am in a good mood, I actually read the emails. Once in a while, I actually respond. On rare occasions, I actually try to answer the questions. The following is one actual interchange between me and an adoring fan... I took out all of the embarrassing stuff. 

The question


A recent question came up for which I didn't have an answer. I immediately thought of you so here goes. Do you have any experience with reading florescent colors? I’m not really sure what types of devices might be out there (if any) for reading those really wild colors used in some printing and certainly more and more in fabrics.

I’m not sure how a profile would even be put together because the standard tools (100% maximum colors) don’t seem to apply. 

Do you know of any devices out there targeted toward florescent colors?



The response

Dear Mike,

This is a very difficult question. Well... the question is easy, but the answer is hard!

I have three different answers:

1. Scientific answer

The characterization of a fluorescent color is not a one-dimensional spectrum, but a two-dimensional one. You illuminate the same with a "monochromator", illuminating it at one wavelength at a time. For each wavelength in, you measure a full spectrum out. This gives you what is called a Donaldson matrix. From this you can predict the CIELAB value for any type of illumination possible.

You pretty much have to build your own spectro if you want to do this. I am sure you could do a decent job of $50K if you had a metrologist to help you. NIST has one and Avian has one. The equipment pretty much needs a technician in a lab to run it, it takes tens of minutes to make a measurement, and you probably have to write special software to interpret the results.
Clearly this is not a good production solution!!

2. Industry answer to a narrower problem

The print industry has faced a similar problem, but limited to one fluorescent pigment. Paper manufacturers currently add stilbene to virtually all paper. It's a cheap way to make paper that is whiter than white, or even just not dingy. Stilbene absorbs UV light and re-emits it in the blue region so as to undo the natural yellow or brown color of paper.

The standards groups huddled together and came out with a solution that is to standardize the UV content in viewing booths and in spectros. Coincidentally, I recently blogged on that topic.
This handles one fluorescent whitener. It was not intended for DayGlo orange or neon green.

3. Perhaps a practical solution

The color of a fluorescent sample will vary depending on the spectrum that it is illuminated with. It will look different under daylight versus incandescent lighting. But, so long as you restrict yourself to one illumination spectrum, there might not be a problem. If the printer and print buyer can agree that they will visually evaluate under a specific illumination and that they will measure with an instrument that has that same illumination, then everything should work.

[Note: Instruments have settings for different illuminants, such as D50, A, F11... This does not change the illumination, just the calculation afterwards. I blogged about that recently, too.]

The tough part (you would think) would be to find a viewing booth that uses the same illumination as a spectro. But actually, that's not so hard because of #2. Theoretically, you should be able to use a relatively new viewing booth (one that complies with the M1 condition in ISO 3664:2009), and a relatively new spectro (one which also complies with the M1 condition, but in ISO 13655:2009). All the stuff with then provide D50 illumination.

In practice, this may not be as easy as it sounds. One issue is that the instruments and viewing booths may simulate D50 in such a way as to have the correct numbers on paper with FWAs - on stilbene - but with somewhat different spectra.

I would suggest sticking just to one make and model of spectro, and one make and model of viewing booth. Unfortunately, I can't tell you which viewing booth and spectro will agree with each other. The vendors don't readily share this information.

I think at the very least, the practical solution might be to do away with any visual matching, and rely completely on measurement. You would measure a color with one of the M1 instruments (XRite eXact, Konica Minolta FD-7, Techkon SpectroDens, or Barbieri SpectroPad) and set that as the standard numbers and instrument.

The other part of you question has to do with profiling... That's a big "yikes"!  I am going to guess that almost all the fluorescent colors are well outside the gamut of all proofing devices, so, what good does it do to proof it?!? The best you could possibly do is use a softproof, and adjust something or other. I think you would scale the spectra of the whole profile to make sure that there are no points where the spectra goes above 100%.



After I answered this email, I contacted DayGlo to see what they do for quality control. Here is what they said:

Visual light source is Daylight North Illumination (D65).
We measure the Color with a X-Rite Color i5 colorimeter.
We record the L*, a*, b*, DEcmc and De* for each color.
The measurements are done on the primary color test.
These can be either drawdowns or prints depending on the product.

So... they chose the practical solution.

Wednesday, November 12, 2014

How many D65's are there in a 2 degree observer?

The wonderful thing about standards is that there are so many to choose from. Today's rant on the topic is aimed at the various ways to calculate CIELAB values. You may have thought that a CIELAB is a CIELAB is a CIELAB, but there are at least eleventy-two different ways to combine the "illuminant" and the "observer", each giving you somewhat different values for L*, a*, and b*. 

This may seem dumb. After all, an object has a single color and that's it, right? (And by the way, that color is whatever my wife says it is. She just told me to find her peachy-pink scarf. Hilarity ensued.) But it's not as simple as that. The color of an object is subtly dependent on two things: the light that is shining on it, and, yes, it does depend on size.

Nine different delicious ways to prepare SeaLab


When I ask a simple question like "what color is a banana?" the answer is obvious to any five-year old. It's yellow. Under normal conditions, the banana will retain this color, but put on deep red sunglasses (or equivalently, look at the banana under red light), and the banana looks red. Put on blue sunglasses and the banana suddenly looks like the kind of banana that my wife would expect me to eat.

Colorful lighting can enhance the look of food

The depraved idea that an object has a specific color independent of the illumination is built right into our language. One should really ask "what color does the banana appear under a such-and-such lighting?"  The reason we think of an object having a color apart from the effect of the light shining on it is that the eye and brain have conspired to make us believe in color constancy.

How is this trick accomplished?  First, the eye and brain together do a remarkably good job of auto-ranging. If I am reading a magazine as I walk from indoors to outdoors, I may blink a few times, but it never occurs to me that the intensity of light hitting my eye just changed by many orders of magnitude.

This auto-ranging is performed separately for each set of cones in the eye, which is an important little detail needed to preserve color constancy. When I go from warm incandescent light to cool fluorescent light, the relative amount of light at the blue end changes by almost a factor of ten. Do you notice that? Huh?  And why didn't you notice it? Auto-ranging, my friend.

Here's an example of the conspiracy which has become famously known as the "Jennifer Aniston has way different RGB values for her skin tone from picture to picture" effect. Someone wrote a blog post about the color of her forehead. Something about how the eye/brain tries to get us to not see the glaring difference in color. The blog post went viral, so now everyone is talking about the "Jennifer Aniston has way different RGB values for her skin tone from picture to picture" effect. 

The reason that these pictures of my heartthrob all appear "normal", is a devious trick in the brain that goes beyond the auto-ranging feature. Somehow or other, the lower levels of the brain have this concept of white point. Everything that we perceive is not thought of in absolute terms, but rather, in relative terms to some color that we have decided is "white". The brain somehow has established a different white point for each of the six gorgeous pictures.

Here is another cute trick to illustrate this devious trick. Go to your recycle bin and pull out a newspaper, or some similar paper. Look at it for a while, against a dark background. What color do you perceive the paper to be? I am guessing that you see white.

Now, unbeknowst to your eye and brain, set a piece of printer paper or a glossy magazine next to the newspaper. As quick as you can say "John the Math guy is a genius", the newspaper will take on a dingy tone. A new white point! That lower visual system is a tricky little critter, I tell you!

So, part of the process of computing a CIELAB value is to factor in the illumination in kinda the same way as the eye.

Object size

This is totally non-obvious, but objects change color as their size changes. Or, to be more specific, objects change color depending on the angular size of the object. Well, actually, to be even more specific, the spectral response of the eye depends upon the position on the retina. The area of the retina that collects light from within 2 degrees of "dead on" has one spectral response. Once you are talking about objects extending to 10 degrees, the spectral response changes slightly.

How big is 2 degrees and 10 degrees? At arm's length, 2 degrees is about the size of your thumb, and 10 degrees is about the width of your fist.

Rule of thumb: If you can smell his breath,
the beautiful stripes are likely to be subtending an arc of at least 10 degrees

Going from spectrum to CIELAB

Now, on to explain how the computation of CIELAB values are computed in order to take into account all these crazy effects.

The first step is to measure the spectrum with the spectrophotometer. This is just reflectance as a function of wavelength -- what percentage of 570 nm (yellow-green) light will reflect from the sample?  This by itself has nothing to do with the light source; nothing to do with either the light source of the spectro or the light source that has been specified for the CIELAB values.

One possible confusion... The spectrum of the sample is not D50 or D65 or F11 or illuminant A. It has been sanitized to remove all traces of the illumination. When you flip the switch on your spectro to make it report D75 instead of F3, you don't change the light that hits the sample.

Once you have the spectrum, the next step is to apply the spectral curve of the light source... D65 for example. The purpose of this is to simulate how much light at each wavelength is available to reflect. For example, if your light source is incandescent (illuminant A), then the amount of light at the blue end is much smaller than at the red end. The spectral curve for illuminant A will dampen down the blue end. The calculation that goes on inside the spectro accomplishes this by applying one of the standard illuminants from CIE 15, ASTM 308, or ISO 13655.

This monarch is blissfully unaware of all the math that goes into color

The third step is to apply one of the sets of curves that approximate the spectral response of human eye. These are the tristimulus functions. This will reduce that spectrum of 15 or 31 or 36 numbers down to three (X, Y, and Z) which represent the responses of the three cones in the eye. Here, we have two choices: the 2 degree (thumb-sized or smaller) or the 10 degree observer (fist-sized or bigger).

(For the persnickety reader, I should admit something. The XYZ curves are not actually the spectral response of the three cones. Those responses are called LMS. It's pretty simple arithmetic to go between the two, but the exact values of the constants weren't known back in 1931 when the 2 degree observer became a standard.)

At this point, we have approximated the signal that travels along the optic nerve to the brain. The next step is to approximate the math that the lower level of the brain does. We translate the XYZ values into L*a*b* values to get this.

Well, sorta. There is some further math that the brain does. At the cognitive level, we don't think in terms of a* and b*, but rather in terms of chroma and hue. So, to get numbers that intuitively make sense, we convert from L*a*b* to LC*h. (This conversion can go either way - they represent the same information.)

Usage notes

So, we got us a couple of plethoras full of combinations of ways to multiply stuff together to convert the spectrum of a sample to a color value. How is a color metrologist to decide which delightful combination to choose?

Steve Carrel demonstrates the proper usage of CIE standard illuminants

Here is a list of the illuminants that are mentioned in ISO 15-4:2004. The silly four digit numbers like 2856 and 6800, are called correlated color temperature. A smaller number refers to a warmer (redder) color, a higher number refers to a cooler temperature (bluer). The silly two digit numbers are abbreviations of the four digit numbers; D65 actually means D 6500 K.

    Illuminant A - 2856 K, kinda looks like an incandescent light bulb.

    Illuminant B - 4900 K, deprecated, which means "not recommended anymore".

    Illuminant C - 6800 K, kinda like daylight, but in the shade.

    Illuminants - daylight simulations D50, D55, D65, D76.

    Illuminants FL1, FL2, ... FL6 - standard fluorescent lights.
    Illuminants  FL7, FL8, and FL9 - broad band fluorescent lights.
    Illuminants FL10, FL11, FL12 - narrow band fluorescent lights.

I am not an arithmetician, but I count 18 different illuminants.

There are two "observers", the 2 degree and the 10 degree, that have been defined. That brings us to a total of 36 combinations.

There is a general recommendation from color scientists to use D65 for computing CIELAB values, but the choice is industry specific. The graphic arts community (of which I am a card carrying member) has chosen D50 to be its standard illuminant. Why? Maybe D50 was taken as a compromise between theoretical daylight (D65) and theoretical living room light (A). Personally, I think the D50 lobby bribed someone on the standards committee, but I can't prove anything.

Similarly, the standard in the graphic arts for the XYZ function is the 2 degree observer. Why? Areas of constant color in the Victoria's Secret catalog are usually pretty small. Well, unless the colors are banners and background, in which case, who cares about the color?

So, if you are doing printing, then the ISO Technical Committee 130 has decided on D50/2 (simulation of the color under D50 lighting, with a colored area being about the size of your thumb).

But what about packaging printing?

Imagine that Jamal is standing in front of that big screen TV box at Best Buy. He is making that spur of the moment decision about spending three month's salary on that TV that's going to make him not care whether the Packers win. The color of that box darn well better look good. And the colors probably subtend an arc greater than 2 degrees. Oh... and the lighting in the store? Probably not D50.

So, a good case could be made that the color of this package should be computed with the 10 degree observer and probably with one of the F illuminants. Ideally, it would be computed with the F illuminant that is in the store. This is what contracts are for. If I was Mr. Sony, I would specify that Big Screen Package Printing use the 10 degree observer for their color computation.

But, yet again...

When it comes down to it, it generally doesn't make a huge difference if you specify D50/2 or D65/10. It's a much bigger deal to make sure that everyone has agreed on which combination is to be used.