Color appearance under LED illumination :

The visual judgment of observers

Françoise VIÉNOT*, Elodie MAHLER*, Jean-Jacques EZRATI**, Clotilde BOUST**, Albane RAMBAUD*** and Alain BRICOUNE****
 
*Centre de recherche sur la conservation des collections, MNHN, CNRS, MCC, Paris, France
**Centre de recherche et de restauration des musées de France, Paris, France
***Ecole nationale des travaux publics de l'état, Vaulx-en-Velin, France
****LedToLite, France

ＡＢＳＴＲＡＣＴ
The advantages of LED lighting are discussed. We report experiments where observers graded the quality of several light-emitting diode (LED) illuminations. One experiment is based on color discrimination and two are based on the judgement of appearance. We conclude that clusters with red, green, blue, and/or amber LEDs impair color discrimination, although they seem to locally enhance the color gamut. We also conclude that LED clusters that include white LEDs and a few correcting color LED may render colors faithfully.

KEYWORDS: white LED, lighting, color discrimination, color appearance

Introduction: Advantages of LED for lighting

Light-emitting diodes (LED) constitute an emergent technology. It has been possible to produce white light for only a few years. Today, LEDs are available in a variety of colors and offer positive characteristics: compactness, adjustable intensity, long lifetime and potential energy saving. The emitted light is deprived of ultraviolet radiation and infra-red radiation, which improves efficiency and safety.
Various LED configurations allow to shape the light spectrum. By principle, the spectral power distribution of the light emitted by LEDs is narrowly tuned, and the light strongly colored.
To produce white light with LEDs, two ways are mainly employed. Either the manufacturer assembles several diodes, typically red, green and blue ones. The mixture of the three radiations produces a white light. Or he manufactures a white diode from a blue diode covered with one or several phosphors that collect a part of the blue radiation and returns, by fluorescence, a radiation of higher wavelength in the yellow part of the spectrum. Again, the addition of blue and yellow lights produces a white light. Whatever the solution, the spectrum of this artificial light is definitely different from the spectrum of the natural light, which can pose problem.
Compared with traditional technologies, LEDs bring a great flexibility in the choice of the color of illumination. According to the process of assembly selected, colorimetric calculations allow to calculate the intensity of the LEDs to be assembled. It is also possible to enrich the light and optimize the final color rendering if a large number of different LEDs is assembled.
Today, the production of white light with LEDs is a controlled technology. Here we question the quality of the white light thus emitted.

The quality of light

White light and the color temperature

What is "white light"? The light is white by reference to daylight under which mankind has evolved and which delivers a complete spectrum to which we are accustomed. Admittedly daylight is changing, according to the hour of the day, the state of the sky, the sunning, and the season. But its margins of variation are strictly limited and its spectrum does not practically deviate from an average spectral model.
It follows that the natural light can appear gilded or pinkish, and in this case, we judge it as “warm”, or it can appear “neutral”, or even very slightly bluish, and in this case we judge it as "cool". To precisely characterize the color of a white light, lighting engineers have introduced the scale of "color temperature": Warm environments are characterized by low color temperatures (around 3000 K), cool environments are characterized by high color temperatures (around 6500 K). This scale reflects only the color of the light, and does not inform of anything on its spectrum.

Light quality and the color rendering index

What is "light quality"? Quality can be considered according to various scenarios.
It can be a question of guaranteeing the fidelity of the colors of objects, as in colorant and dye industry.
It can be a question of ensuring the fine discrimination of the nuances of color, as the ones of a masterpiece in a museum.
It can be a question of restoring a natural appearance, as in cosmetics.
It can be a question of bringing comfort at night fall, or visual performance in an office environment, or giving a pleasant aspect to the environment, or of any general appraisal.
The color rendering index (CRI) informs suitably about fidelity. The International Commission on Illumination (CIE) has defined the color rendering index which allows, for each source of light, to assess its quality in reference to daylight [1]. This color rendering index is obtained by comparison of the color of a collection of colored samples lit by the source to test, with their color under a source of reference. The CRI reaches value 100 for the source of reference.
A collection of eight samples specified by their spectral reflectance yields the general Color Rendering Index Ra. Six additional samples specified by their spectral reflectance yield particular Color Rendering Indices. In this paper, we note Ra14 the average of the CRIs of the 14 samples.
In the past, improvements have been proposed [2,3], and in 2006, a technical committee has been established by the CIE to investigate new developments.

The judgment of the observer

Recently, the CIE reported visual experiments and simulations and concluded that the present “CIE CRI is generally not applicable to predict the color rendering rank order of a set of light sources when white LED light sources are involved in this set” [4]. Precisely, the calculations have shown that the choice of test samples was critical since the CRI could achieve high scores for some samples and poor scores for others. The experiments have also shown that the visual impression of colorfulness could be as relevant as color fidelity to assess the quality of lighting, and that this parameter was ignored in the calculation of the current CIE CRI [3].
Let us recall that, in the visual system, information processing is parallel and obeys a hierarchic scheme. Details and color differences are processed locally in the retina. Integration of color and appearance takes place in the brain and concerns large areas of the visual field.
Visually, it is the judgement of the observer which will decide upon the quality.

Experimental studies

Objective of the studies

Here, we have collected the judgment of observers, using methods based on color discrimination and on color appearance. We have compared the light quality of various LED configurations with that of conventional illumination. Parts of the results have already been presented in another conference [6].
Various LED configurations were appraised by observers in real situations.
The rationale is unique. It consists of comparing the color of a collection of samples illuminated by the test light with their color under a reference light. The observer is invited to judge small differences between samples, or to assess a value to a color appearance attribute.
Several LED clusters were housed in a light booth that allowed normal viewing conditions. Practically, we assembled:
- a cluster of red, green and blue LEDs (RGB),
- a cluster of red, green, blue and amber LEDs (RGBA),
- a cluster of two phosphor cool white plus red LEDs (WWR),
- an "enriched" mixture of cool white, warm white, red, green, blue and amber LEDs (WWRGBA).
The source of reference was a regular filtered tungsten-halogen lamp, emitting at 4000 K, a light close to that of the blackbody or of the direct light of the sun.

Figure １. Spectral power distribution of the light emitted by the sources used in the color discrimination experiment, measured using the Minolta CS-1000 spectroradiometer with 5 nm bandwidth. Similar curves were available for the sources used in the color appearance experiment.

All spectra were measured using a Minolta CS-1000 spectroradiometer (Fig. 1). All lights had the same correlated color temperature Tcp. They produced the same illuminance E. In the case where more than three LEDs were used, their intensity was adjusted in order to optimize the CRI (Table 1). The general and special color rendering indices (CRI) Ra and Ra14 were calculated according to the Test-Color Method as recommended by CIE [1].

Table1  Illuminance E (lx), color specification x, y, color temperature Tcp (K) and color rendering index Ra of the light emitted by the sources used in the color appearance experiment. Similar values were available for the sources used in the color discrimination experiment.

Color discrimination

Hue discrimination

Details of the hue discrimination experiment are available in a previously published manuscript [6, 7]
We prepared 32 color caps, equally distributed along a color circle in the CIELAB color space (L* = 78.54 ± 0.69, C*ab = 15.05 ± 2.16, ΔE*ab = 3.02 ± 0.55 CIELAB units).
Forty observers having normal color vision participated in the testing. Each observer was invited to order the caps under every illumination. When performing the test, an observer could either succeed in ordering the caps, or fail that is to say make permutations of adjacent or non-adjacent caps. The hypothesis is that if many observers fail the test, it probably reflects a defective quality of the illumination.

Figure 2. Number of observers, out of 40, who failed the discrimination test under a given illumination.

For each illumination, we counted the number of observers who failed to order the samples (Fig. 2). The RGB and the RGBA clusters produce about twice as many errors as the WR cluster, the WWRGBA cluster and the Solux incandescent lamp (Fig. 2). Further, errors under RGB illumination often consist of permutations between non-adjacent caps.
In terms of perceptible colour difference, it corresponds to an increase of 3% for the Solux, the WWRGBA and the WR illumination, of 5% for the RGBA illumination and 7% for the RGB illumination.
Specifically, the errors are not evenly distributed along the color circle. They are numerous around red and bluish green shades. We will come back to this point in the discussion.

Color appearance

Colorfulness

During the color discrimination experiment, the observers orally reported an increase of colorfulness under RGB illumination. For this reason, we carried out an experiment to verify and quantify the increase of colorfulness.
We used a collection of 38 samples from the Natural Color System (NCS) atlas, at Blackness equal to 0.5 and Chromaticness equal to 20. All hues were represented, except for two samples that are not manufactured by NCS.
Twenty normal color observers participated in the experiment. They were invited to rate the apparent colorfulness of isolated samples on a 0-10 scale. Zero was supposed to be used for the neutral grey. The observers were instructed that the rating value should increase as the colorfulness appears to increase.

Figure 3. Average colorfulness rating of 20 observers.

The results show that the apparent scale is expanded with RGB and RGBA illumination for almost all observers. When individual results are summed and averaged, yellowish-red and bluish-green samples tend to get more colorful under RGB or RGBA illumination than under WWR, WWRGBA or Solux illumination (Fig. 3). We note that the expansion of colorfulness and the impairment of discrimination are located around the same reddish and bluish-green shades as the ones around which the discrimination errors occur.

Hue naming

The following experiment was planned to comfort the change of appearance under various illuminations.
The observers were presented with the 38 samples of the NCS samples and asked to name the hue of individual samples. They were instructed to use two out of the four elementary hue names: Red, Yellow, Green, Blue. The first named hue was considered as the dominant hue of the sample. Therefore, the hue circle is apparently divided in 8 segments according to the observer’s judgment. Each segment corresponds to a named dominant hue and a named secondary hue. This is the hue combination “Major/Minor Hue” categories described by Harrar [8]. For instance, reddish blues are categorized as “Blue-Red”.
Any change of appearance between illuminations is traceable back to a shift of the border between adjacent Major/Minor Hues categories and is quantified by averaging between observers the perceived hue of the samples.
Twenty color normal observers took part in the experiment. During one session, one observer was asked to examine the series of samples, under the five different illuminants.
Results show that under RGB illumination, hues are apparently attracted toward the yellowish-red or the greenish-blue categories (Table. 2).

Table 2 Classification of NCS samples (left-column) with respect to “Major/Minor Hue” categories.

Discussion

Predictions of visual results from the CIE color rendering index

Among the particular LED clusters that we have tested, only those including white LEDs yield acceptable CIE CRIs larger than 80 (Table 1).
When we examine the results of the color discrimination experiment (Fig. 2), we conclude that the CIE color rendering index is globally a reasonable predictor of the color discrimination efficiency of the light source.
However, the CRI is only a single-number index that can conceal considerable color rendering variations between samples.

Predictions from CIELAB

There is a considerable advantage to examine how the results of all samples are distributed.
When we plot on an CIELAB a*b* diagram the colors of the caps used in the discrimination experiment, we can see that, under RGB illumination, the color circle is distorted and elongated, but the intervals between neighboring caps are reduced at the apices of the elliptical distribution, where the red and the bluish green shades are located (Fig. 4).

Figure 4. Color specification in CIELAB of the collection of 32 samples used in the discrimination experiment.

So the increase of errors corresponds to the lowering of the color differences between the caps.

Predictions from CIECAM02

We also computed the “Chroma” value of the 38 NCS color samples, using the CIECAM02 color appearance model.
The comparison between the experimental results and the CIECAM02 [9] predictions shows that the apparent colorfulness appraised by the observers and the chroma computed using CIECAM02 follow the same pattern of variation over the color circle. Both approaches also show an increase of colorfulness (chroma) for yellowish-red and bluish-green samples under the RGB illumination (Fig. 5).

Figure 5 Comparison between colorfulness experimental results and CIECAM02 Chroma predictions.

A paradox

When considering every sample, we arrive at a paradox. With RGB or RGBA LED configurations, the increase of colorfulness was noticed around the reddish and bluish-green shades, precisely where the observers made the largest number of errors. So the apparent colorfulness of colored surfaces of a scene can be increased, precisely where the fine discrimination of the colors is impoverished.
From our experiment and others’ [4], it seems that the judgment of color appearance might not be predicted by the color rendering index. Besides, RGB clusters produce the highest error rates [10]. Further, from our experiments, colorfulness enhancement and color discrimination impairment seem to be conflicting.

A color vision model

The fact that the visual impression could be described by two factors, one associated with colorfulness, the second one with color fidelity had already been mentioned by Nakano et al. [5]
From our experiments, we observe that the colorfulness increases precisely for the shades where the color discrimination is impaired.
Indeed, RGB illumination physically tunes all stimuli around three spectral peaks, while it impoverishes the spectral signal between the peaks. Therefore, the purity of the stimulus increases, resulting in saturation and colorfulness enhancement. The spectral tuning also explains that colored materials that are discriminated under a full spectrum illumination such as the Solux or the WWRGBA illumination could look similar under spectrally tuned illumination.
However, the spectral tuning of the stimulus around three regions of the spectrum does not explain that the colorfulness is enhanced around two regions only of the color circle. To our opinion, this indicates that the colorfulness enhancement cannot be predicted from the spectral distribution of the stimulus only, without considering the visual mechanisms. Let us recall that color appearance results from a visual process which begins with the absorption of photons by three families of cones. Then the signals from the cones are linearly combined through a luminance pathway and two chromatic contrast pathways. Finally, the chromatic attributes of color, such as hue and saturation, are extracted from the signals of the two chromatic contrast pathways. Any spectral change of the light entering the cones modifies the balance of the two chromatic contrast pathways. The resulting change of appearance is usually two-dimensional.
Rather than thinking in terms of tuning the stimulus, thinking in terms of modifying the spectral sensitivity of the cones would better explain the modification of appearance.

Conclusion

We conclude from our visual experiments that for the faithful color rendering of the environment and for the fine discrimination of shades, the configurations RGB or RGBA that we have used should be avoided.
Surprisingly, these configurations seem to enhance the colorfulness around yellowish-red and bluish-green hues. So the apparent expansion of the color gamut is at the expense of color discrimination.
Conversely, "white" LED configurations enriched with a few correcting color LEDs provide a much more satisfactory illumination.
Finally, color discrimination is a major issue. Any new color rendering quotation should take into account color discrimination. Further, information should be available that fully describe the color effects produced by the light.

Acknowledgments

We acknowledge for his comments on the manuscript and the volunteer observers.

References

(1) CIE. Method of Measuring and Specifying Colour Rendering Properties of Light Sources. (1995) CIE 13.3.
(2) CIE. Research note: Colour rendering, TC 1-33 closing remarks. (1999) CIE 135/2 10-22.
(3) W. Davis, Y. Ohno. Proc. Solid State Lighting V, SPIE (2005) Vol. 5941:283-290.
(4) CIE. Color rendering of white LED light sources (2007) CIE Technical report 177:2007.
(5) Y. Nakano, H. Tahara, K. Suehara, J. Kohda, T. Yano, AIC Colour 05 (2005) 1625-1628.
(6) F. Viénot , J.-J. Ezrati, C. Boust, E. Mahler, CIE 26th session (2007) D1:22-25.
(7) E. Mahler, J.-J. Ezrati, F. Viénot , Submitted to Color Research and Application.
(8) M. Harrar, G. Baillet, B. Bourdoncle, H. Brettel, K. O’Regan, F. Viénot, AIC Color 05 (2005) 643-646.
(9) CIE, A color appearance model for color management systems: CIECAM02. (2004) CIE 159-2004
(10) F. Szabó, J. Schanda, P. Bodrogi, E. Radkov. Proc. CIE 26th Session (2007) D1:18-21.

LUX magazine N°245 Nov-December 2007