AI and the Internet Need to Learn Color Rendering

Recently a colleague of mine from the IES Color Committee asked of one of the new AIs “Tell me about color rendering.” If we assume that the answer was formulated based on information about color rendering that is available on the internet we have a problem because the answer the AI returned was almost completely wrong. That strongly suggests that most individuals and manufacturers posting information about color rendering don’t fully understand color rendering and are perpetuating outdated or incorrect information. Here’s what the AI had to say, and what’s wrong with it’s response.

Color rendering is a measure of how well a light source reveals the true colors of objects. (1) A light source with a high CRI (Color Rendering Index) will show objects in their natural colors (2), while a light source with a low CRI will make objects look dull or washed out (3). The CRI is a scale from 0 to 100 (4), with 100 being the perfect score (5). A light source with a CRI of 100 will show objects in their true colors (1) while a light source with a CRI if 0 will make all objects look the same color (6).

The CRI (7) is important for applications where it is important to see objects in their true colors (1), such as art galleries, hospitals, and retail stores (8). It is also important for applications where it is important to see subtle variations in color such as fashion design and photography (9).

There are a few things that can affect the CRI of a light source (10). The type of light source (incandescent, fluorescent, LED, etc.) will affect the CRI (11). The color temperature of the light source will also affect the CRI (12). Light sources with a higher color temperature will have a higher CRI (12).

The CRI (7) is a valuable tool for choosing the right light source for your application. By understanding the CRI, you can choose a light source that will show your objects in their true colors (1) and help you achiever the desired look and feel for your space.

Objects don’t have “true” colors, they have apparent colors. The color of an object is determined by the spectrum of light striking it, it’s reflective properties, and the resulting light directed toward the observer. An object’s reflective properties don’t change, but the spectrum of light changes every time the light source changes. Changing the spectrum may, therefore, change the color appearance of the object, hence apparent color.
While there is a correlation between fidelity and naturalness, they are not the same thing. CRI measures fidelity, i.e. how well a given light source matches the color rendering of a defined reference light source.
One of the problems with single measure metrics like CRI is that there’s a lot of information that isn’t conveyed. As CRI values drop, the only thing conveyed is that the match to the reference light source is worse. A worse match, however, doesn’t mean colors are made dull. It could be they are increased in saturation since both deviations from the reference are equally penalized. That’s the advantage of TM-30. As Rf decreases we can see why by looking at Rg and some of the other 147 measures.
CRI can have negative values. TM-30 Rf is calculated so that 0 is the lowest value.
100 is the highest value. It’s dangerous to call it “perfect” though as that implies that high fidelity is the only color rendering goal, which it isn’t. TM-30 provides information for the color rendering goals of preference and vividness, and may include more in the future.
A CRI of 0 will certainly make nearly all colors look terrible and very similar, but not all the same.
CRI isn’t a proper noun, and shouldn’t be preceded by “the”.
There are strong arguments for emphasizing preference over fidelity in many applications, including retail. Again, fidelity isn’t the only color rendering goal, although it is the only one CRI measures.
Research shows that high fidelity isn’t necessarily the best spectrum for detecting color difference. Additional research is needed, but the IES may eventually add a color difference metric to TM-30.
Only one thing affects CRI value – the spectrum of the light source.
This is true because different light producing technologies have similar quirks in their spectra. Those similarities can lead us to blanket statements such as “all fluorescents are green” which are not true for all products. Again, the individual light source’s spectrum determines everything.
A common misconception, but not true at all. Not in the slightest. CCT and CRI are separate metrics.

Fixing CRI, But Not Really

A couple of weeks ago the Global Lighting Association (GLA) published Application of CIE 13.3-1995 with Associated CRI-based Color Rendition Properties. It proposes TM-30 like metrics to supplement CRI R_a. Specifically, it proposes a color gamut index, G_a, that is similar to TM-30’s R_gand a set of chroma indices, C_i, similar to TM-30’s R_ch,hj. At first glance I can see some specifiers getting excited about this. Since it’s based on CRI it’s already somewhat familiar so it should be easier to learn. But…

The first part is the problem – it’s based on CRI, which has well known and well documented problems and shortcomings. As described in The Lighting Handbook, 10^thEd.and IES DG-1 Color and Illumination, they include:

Averaging the Color Shifts. CRI is computed by averaging the color shifts of the eight color samples. A light source can render one sample very poorly and still achieve an acceptable score. The Chroma Indices attempt to resolve this, but there’s a problem with the eight color samples, which is next on our list.

Test Color Samples. The eight color samples are all of moderate saturation, so saturated colors can be rendered poorly even when CRI is high. More importantly, the eight color samples are A) not based on any real-world objects B) don’t adequately cover the visible spectrum. The latter is significant because it means that some wavelengths play an outsized role in determining CRI. Manufacturers can use this to optimize (or cheat) their spectrum to achieve a higher CRI than visual inspection would warrant. This is why TM-30 uses 99 color samples drawn from real world objects.

Color Space. CRI is calculated in the CIE 1964 color space, which is no longer recommended for any other use because it is outdated.

Penalties for all Chromaticity Shifts. As a fidelity metric, CRI penalizes all chromaticity shifts even though research has shown that certain increases in chroma are preferred. Again, it seems that C_iis intended to address this, but it doesn’t resolve the other problems with CRI.

Chromatic Adaptation. The chromatic adaptation transform used in CRI has been shown to perform poorly and is no longer recommended for any application.

CRI served the industry (relatively) well, but its time is over. Layering new calculations on top of CRI’s flawed foundation doesn’t make it better. I know TM-30 can be tough. But it works. It tells us what we want to know at the level of detail we need for a given project, and it’s accurate. The IES has moved on and the CIE seems to be leaning in that direction as well. So should you!

TM-30 and Daylight

An architect recently emailed me asking if it was possible to use TM-30 metrics with daylight. My short answer was something like, “I suppose you could, but why would you?”

The long answer is that all of the TM-30 measurements (Rf, Rg, the 16 chroma shifts, the 16 hue shifts) are relative – comparing the light source in question with the reference light source. With daylight, you’d be comparing the daylight SPD you captured at a moment in time with the CIE definition of daylight at the same CCT. On an average day I doubt that any of the measurements would deviate from 100 by more than a couple of points. So, using TM-30 (or CRI Ra) is like measuring a ruler with another ruler. You’re essentially comparing one thing to a definition of itself.

R.I.P. CRI

It’s been a little over two years since the IES released TM-30-15 IES Method for Evaluating Light Source Color Rendition. In that time TM-30 has seen growing support in the industry and a growing body of evidence for its accuracy and usefulness. We’ve nearly reached the moment when we can all agree that it’s time to retire CRI and fully adopt a modern, accurate system of measuring and describing the color rendering of light sources. What’s wrong with CRI? Quite a bit, so if you’re not up to date on the issue here’s an overview.

In 1948 The CIE first recommended a color rendering index based on a method developed in 1937. The 1937 method is a fidelity metric (that is, it compares a test light source to a reference light source) that divides the spectrum into eight bands and compares each band to a full spectrum radiator. In 1965 the CIE finally adopted CIE 13-1965 Recommended method of measuring and specifying color rendering properties of light sources, based on a test color sample method, what today we call CRI R_a or just CRI. From the start it was apparent that there were problems. In 1967 a committee was established to correct for adaptive color shift. Other problems were uncovered, and in 1974 a formal update was published. Errors were uncovered in the 1974 edition, resulting in a third version in 1994, which is the version we use today.

So far, so good. Errors are discovered in the method used and are eventually corrected, so what’s the fuss? The fuss is that the corrections were minor compared to the scope of the errors, and 23 years after the last correction we still don’t have an accurate, up to date system. In the early 1990s a proposal to update the formula and test color samples failed to gain consensus. Two subsequent attempts to improve the metric also closed without adoption. The current problems, as described in the 2011 IES Lighting Handbook, 10th Edition include:

Averaging the color shifts of the eight test colors says nothing about the rendering of any single sample. A large error in one color can be masked by accurate rendering of the other samples.
The test color samples are all of moderate saturation so the index doesn’t reveal color shifts in saturated colors. In addition, the test colors are not evenly distributed through the color space or the spectrum, so light source spectra can be engineered to score higher than visual observation would indicate.
The color space used, the 1964 UCS chromaticity diagram, is no longer recommended for any other use.
All chroma shifts are penalized, even though research shows that moderately increasing chroma is desirable in many applications.
The chromatic adaptation used has been shown to perform poorly and is no longer recommended for any other use.
A single number index gives no information about the direction or extent of color shift for any particular color or color range.

Why haven’t these problems been corrected in the past 23 years? I’m told that there are two issues. The small issue is that competing scientific interests on the committee advocate new metrics that they’ve developed as a replacement or supplement to CRI. The larger problem is that manufacturers on the committee don’t want to see any changes that would reduce the CRI of any of their lamps. From their perspective, it’s better to have a high score on an inaccurate test than a low score on an accurate one. It seems that internal politics has been preventing updates, corrections, and improvements.

Although many other color rendering metrics have been proposed over the years, none has been adopted by CIE, which has the most significant voice on this issue. The result is that the sole internationally accepted metric, which has also been written into product specifications and into codes, is CRI. That began to change in 2015 with the introduction of TM-30. I’ll have more to say about TM-30 in future posts, but for now let’s agree that CRI R_a is broken and CIE is in no hurry to fix it. A better system exists, and our industry should adopt it.

CIE Adopts TM-30 (sort of)

The CIE has issued a statement titled “CIE 2017 Colour Fidelity Index for accurate scientific use” in which they partially adopt the Rf fidelity metric of TM-30. Here are some of the details.

When the IES released TM-30-15 it seemed to be a wake up call for the CIE, who have understood the inaccuracies of CRI but haven’t been able to build the internal consensus to correct them. Two CIE committees soon began work, one on the issue of fidelity and one on other perception related issues. The committee that worked on fidelity (TC 1-90) has decided to adopt TM-30 Rf with a few minor changes to the calculation. They are calling this the CIE 2017 Color Fidelity Index, which will still be described as Rf, and are asking the IES to adopt the same changes to the calculation so that both Rfs are the same.

However, at this time there is no retirement date for CRI, even though they acknowledge that Rf is more accurate. The logic is a bit twisted. Since CRI is used for purposes other than the intended purpose, and since TM-30 doesn’t satisfy those purposes, we’re going to continue to mis-use CRI. Meanwhile, “replacement of the CRI will be a matter of future study and discussion.”

I’ll probably have more to say once I read the entire report. If you want to read the report yourself you can purchase it here.

Design Guide for Color and Illumination

As the co-chair of the IES Color Committee I am delighted (pun intended) to announce the publication of the Design Guide for Color and Illumination. The guide is the result of over five years of work by more than a dozen researchers, engineers, manufacturers, and designers from across the globe. Here’s part of the description on the IES site.

Color can be described using concrete values such as chromaticity coordinates, spectral power distribution, or others discussed later in this guide. However, one’s response to color can be much more personal and emotional—and therefore more difficult to quantify. This guide takes the reader from basic vision and color vocabulary, through methods of measuring and quantifying color, and culminates in the practical use of commercially available white light and colored lights. The definitions, metrics, and references discussed will assist in building a critical understanding of the use and application of color in lighting.

It is probably the best, most thorough discussion of light and color available today. Everyone interested in color, color perception, color rendering, and their relationship to light should read it. It will be available at the IES booth at Lightfair.

CRI Inches Forward

The CIE’s Color Rendition Index (CRI) has long had several known weaknesses including outdated components of the calculations, a limited set of color samples, and standard reporting of only one piece of data (the average color distortion of the first eight colors, known as the General Color Rendering Index or Ra). After more than a decade of stalemate it seems that the IES’s TM-30 has convinced the CIE to make another attempt at updating CRI.

In a recently released position statement the CIE announced that the CRI technical committee (TC 1-90) has taken up the problem again and is expected to write a technical report on a new, improved color fidelity metric that can update CRI before the end of 2016. They will use TM-30 R_f as the basis for their work. Another committee, TC 1-91, will write a report on a color preference metric in the same time period.

This is both good and bad news. CRI is the international standard for measuring and reporting light source color rendering. It is long overdue for an update, and I’m glad to see the CIE working on it. On the downside, the press release makes it sound as though 1) parts of TM-30’s R_f may be incorporated, but the CRI fidelity metric will be a new creation 2) we may see the addition of a color preference metric but TM-30’s gamut metric R_g apparently isn’t being considered. The biggest reason that this is a concern is the amount of time it takes to create the work, gain internal consensus, approve the work, and gain organization approval – all of this before the work can be released to the larger lighting community for consideration. I think it is extremely optimistic to think that the two committees can write meaningful reports on these issues in only one year, especially given the difficulty they’ve had reaching consensus in the past.

I would much rather see the CIE committees study TM-30 and report on its strengths and weaknesses before deciding that it won’t work and they should start over. Adopting or modifying TM-30 can happen much quicker than developing one or more new metrics. TM-30 took three years to develop and the industry shouldn’t have to wait another three years for an updated metric that has the CIE’s approval.

The Best Light?

In class yesterday one of my students, thinking about a project she had recently completed, asked, “What’s the best light for a hair salon?” I’m certain she was hoping I would tell her exactly what lamp technology and/or lamp style to use. Of course, it’s not that simple.

So the class took a detour to talk about the important aspects of light in a hair salon. We narrowed it down to two critical considerations – intensity and color rendering. Intensity is important because the stylist needs to be able to see the details of a head of black hair as well as a head of blonde hair. Intensity is relatively easy to achieve, and the designer has a wide range of lamp technologies, lamp shapes, and fixture types to choose from. Finally, everyone intuitively understands how intensity affects vision. If there’s not enough light one can’t see well enough to work.

Color Rendering is more complicated. All of my students had heard of color rendering, but few of them understood its meaning or use. Color rendering is the ability of a light source to enable us to see object colors. For instance, a light source that produced no red light would do a terrible job of allowing us to judge red apples and we would say it has poor color rendering. Color rendering is measured on the Color Rendering Index (CRI) which compares the light source being tested to incandescent light (for warm light) or to daylight (for cool light). The higher the result, on a range that peaks at 100, the more a light source simulates incandescent or daylight in enabling us to see the colors of illuminated objects.

The best light source, then, is one that produces the desired intensity and has a high CRI. Of course, there’s much, much more to color rendering and to the topic of color in light. The color chapter in Designing Light is about 40 pages, and the IES DG-1 Color and Illumination looks like it will be about 100 pages. It’s critical that lighting designers understand color because it has such a strong affect on people. Color rendering is just one aspect. Color also affects things such as our impressions and perception of a space, circadian rhythms, visual acuity, and the interior designer’s color palette. Those are topics for another post.