Perceived colour fidelity under LEDs with similar R f but different R a

Abstract

Substantial efforts have been made to develop new colour fidelity measures that are better correlated to perceived colour fidelity. Such measures minimize the possibility of increasing rated colour fidelity through spectral optimization without improving perceived colour fidelity. This paper reports a psychophysical experiment that was carefully designed to investigate how light sources with different spectral features affected perceived colour fidelity, and whether the effect can be reflected by the two existing colour fidelity measures (CIE R_a and Illuminating Engineering Society (IES) R_f). Fourteen observers evaluated the colour differences of eight physical colour samples under three pairs of light sources, which had similar IES R_f but different R_a values. IES R_f was found to perform better than CIE R_a.

1. Introduction

The Commission Internationale de l'Eclairage (CIE) General Colour Rendering Index R_a (colloquially, CRI) is intended to characterize the average colour fidelity of a light source.¹ First published in 1965 and unchanged since 1974, it has been an important metric in the lighting community. Lighting manufacturers have used it to develop lighting products; specifiers and users have relied on it to evaluate and compare products; and policy makers have included it in various specifications that have influenced the use of lighting products. CIE R_a, however, has been found to not predict perceived colour fidelity,^2–6 and is incapable of predicting more subjective aspects of colour perception,^2–4,7–12 such as colour naturalness, vividness, or preference. That is, CIE R_a is both inaccurate and incomplete, as acknowledged by CIE.¹³ Its limitations have especially been exposed given the opportunities for spectral engineering afforded by LEDs.

In 2015, the Illuminating Engineering Society of North America (IES) published TM-30-15, IES Method for Evaluating Light Source Colour Rendition, which includes the Fidelity Index (R_f) among dozens of other measures. IES R_f is an updated measure of average colour fidelity, conceptually equivalent to CIE R_a,^14,15 but functionally different due to improvements in the underlying calculation framework.¹⁶ Later, CIE published 224:2017,¹⁷ making minor modifications to IES R_f. The IES has since unified the measures, recently publishing ANSI/IES TM-30-18.¹⁸ IES TM-30-15 R_f has been used in this work, but no differences would occur if IES TM-30-18 R_f/CIE R_f were used instead,¹⁹ as the only notable difference is the scaling factor.

IES R_f specifically addresses the widely documented weaknesses of CIE R_a, such as the colour space non-uniformity and colour sample spectral non-uniformity.^15,16 Nonetheless, some have argued that IES R_f was not significantly different from CIE R_a²⁰ or not worth adopting. Others have since demonstrated large and meaningful differences between IES R_f and CIE R_a, examining real and theoretical light sources,¹⁹ a small number of real light source possibilities,²¹ or numerous real light sources in an experimental setting.^3,4 In particular, Windisch et al.²¹ noted some of the practical effects of these differences related to LED manufacturing and energy efficiency, calling for extra scrutiny on how each measure relates to perception.

Some of the differences between CIE R_a and IES R_f have been examined by psychophysical experiments. For example, several experiments did not find that R_f was better correlated to perceived fidelity (note: the colour samples used in the experiments were not replaced in the calculation),^6,22 while others supported that R_f is a more accurate measure of colour fidelity^5,6,22 and that R_f is more useful in combination with other measures for understanding subjective qualities of illumination.^3,4,7,23,24

Typical white LEDs utilize a “blue”-emitting InGaN chip with a phosphor coating that converts part of short-wavelength “blue” radiation into longer wavelengths. An additional “red” InGaAlP chip is sometimes added to improve the colour rendition of an LED package. Different blue InGaN or red InGaAlP chips have different peak wavelengths, so the spectral power distribution (SPD) of a white LED can be easily adjusted or tuned to achieve a higher colour fidelity score. For example, it was found that a 13-point difference in IES R_f can be produced by simply shifting the peak wavelength of a blue InGaN LED from 450 to 442 nm.²¹ Though the effect of small changes in the SPD of a light source on colour fidelity is important to LED products, only two studies have investigated this issue. Windisch et al.²¹ found through simulations that the blue InGaN peak wavelength had a stronger influence on IES R_f than on CIE R_a, while the red InGaAlP peak wavelength had a stronger influence on CIE R_a than on IES R_f. Smet et al.¹⁶ also performed simulations and similarly found that the CIE U*V*W* colour space, which is used in CIE R_a, has a lower sensitivity to short wavelength shifts than CAM02-UCS. However, it is unknown how the effect suggested by these simulations relate to human perception, outside of the composite dataset used to develop CAM02-UCS. This study was carefully designed to experimentally investigate how small changes in the SPD of a light source, causing little changes to IES R_f but large changes to CIE R_a, affected perceived colour fidelity.

2. Method

2.1 Light source SPDs

Seven different light sources were produced using 14-channel spectrally tunable devices comprised of LEDs with peak wavelengths between 350 and 700 nm. The intensities of the 14 channels were carefully adjusted to produce nearly metameric light sources — one reference and six test light sources — with a nominal correlated colour temperature (CCT) of 3500 K and a D_uv of 0.000.

The reference light source was designed to simulate a 3500 K Planckian radiator with high CRI R_a (97) and IES R_f (96) values. Three pairs of test light sources were designed to have different SPDs, with very similar IES R_f values (between 0.4 and 4.2 points different) but very different CIE R_a values (between 7.7 and 9.4 points different, with one above and one below 80). Specifically, the two light sources in each pair used LED channels with small differences in peak wavelength in one region: “blue” (445 or 450 nm), “green” (510 or 525 nm), or “red” (640 or 660 nm). The SPD of each light source is shown in Figure 1, and colorimetric characteristics are provided in Table 1. These SPDs were measured in situ with a diffuse reflectance standard and a calibrated JETI specbos 1411 spectroradiometer.

Figure 1

Relative spectral power distributions (SPDs) of the light sources measured in the viewing booth with a diffuse reflectance standard and a calibrated JETI specbos 1411 spectroradiometer. They were scaled to an equal lumen output.

Table 1

Colorimetric characteristics of the light sources

	CIE 1931		CCT	D _uv	CRI R_a	IES TM-30-15		R _{a_experiment}	R _{f_experiment}
	x	y	CCT	D _uv	CRI R_a	R _f	R _g	R _{a_experiment}	R _{f_experiment}
Ref	0.4068	0.3899	3461	−0.0007	97.2	95.9	100.7	100.0	100.0
Ref_1	0.4064	0.3907	3476	−0.0003	97.6	96.1	101.0	99.3	99.3
Blue_1	0.4047	0.3893	3502	−0.0005	81.5	77.5	107.4	83.6	78.8
Blue_2	0.4052	0.3914	3507	+0.0003	72.1	75.4	94.7	75.4	74.0
Green_1	0.4067	0.3909	3472	−0.0002	75.4	77.2	109.9	72.1	76.3
Green_2	0.4073	0.3880	3435	−0.0016	84.8	77.6	110.8	85.3	81.9
Red_1	0.4068	0.3900	3462	−0.0006	83.6	79.5	109.0	85.6	84.3
Red_2	0.4055	0.3871	3466	−0.0016	75.9	75.3	111.4	75.1	77.7

In order to better correlate the colour fidelity values of these light sources to the experimental results, custom colour fidelity values, R_{a_experiment} and R_{f_experiment}, were also determined. These measures were calculated by replacing the standardized colour samples and reference illuminant with those used in the experiment, as well as adjusting the scaling factor so that CIE Standard Illuminants F1–F12 had an average colour fidelity value of 75.1 for both measures,^1,14,15 following the scaling procedure used in IES TM-30-15.

2.2 Apparatus and colour samples

Two identical viewing booths, with dimensions of 50 cm (width) × 50 cm (depth) × 60 cm (height) and interiors painted with Munsell N7 spectrally neutral paint, were placed side-by-side. A spectrally tunable LED device (described above) was placed above each booth to provide approximately uniform illumination at 1000 ± 20 lux on the bottom surface, with the equivalence of the booths verified with high dynamic range imaging.

Eight pairs of colour samples (denoted HL1 to HL8) purchased from Sun Chemical were presented to the observers. Figure 2 shows the spectral reflectance distributions measured using a calibrated Xrite Pro 1 spectrophotometer and the CAM02-UCS (a′, b′) coordinates of the samples under CIE Standard Illuminant D₆₅. The samples were placed in the two booths symmetrically, as shown in Figure 3.

Figure 2

Colorimetric characteristics of the eight colour samples (denoted HL1 to HL8). (a) Spectral reflectance distribution (SRD) and (b) chromaticities of the eight colour samples, together with the colour samples used in IES R_f, CRI R_a, and CQS Q_f calculations, under CIE Standard Illuminant D₆₅ in the a′–b′ plane of CAM02-UCS.

Figure 3

Side-by-side viewing booths used in the experiment, with the chin-rest being removed for this photograph. The left booth was always under the reference light source. The greyscale card in the left booth was used to aid the colour difference evaluation (available in colour in online version).

A greyscale card including 10 pairs of samples was placed in the left booth to aid the colour difference evaluation. This technique has been widely used in past studies investigating colour differences.^22,25–27 These 10 pairs had different amounts of colour difference, as conveyed to the observers with a numbered scale printed on the card. The ratings ranged from 1 to 5 with a half point interval, with a greater value for a smaller difference. Note that this scale was the reverse of colour difference value in colorimetry, where a greater values indicates a greater difference.

To minimize variations, the test light sources were always shown in the right booth and the reference light source was always shown in the left booth; this also allowed for stable colour appearance and colour differences for the greyscale card. An additional reference light source, labelled as ‘Ref_1’, was created in the right booth for carrying out a null condition evaluation.

2.3 Observers and experimental procedure

Fourteen Chinese observers (12 males and 2 females) between 20 and 28 years of age (mean = 21.6 years, SD = 2.03 years) participated in the study. All of them had normal colour vision, as tested by the Ishihara Colour Vision Test.

Upon arrival, each observer completed the Ishihara Colour Vision Test and a general information survey, and then the experimenter explained the general procedure. After the observer was escorted to the viewing booths, the ambient illumination in the room was switched off, so that only the booths were illuminated by the experimental LED devices. The observer was seated in front of the booths with his or her chin placed on a chin rest. The observers' sagittal plane was aligned with the panel between the two booths to form a haploscopic viewing condition, so that his or her two eyes were individually adapted to the illumination in each booth.

The observer was asked to look into the booths for 2 minutes, with the reference light source in the left booth and a randomly selected test light source in the right booth, to chromatically adapt to the illumination. During this time, the experimenter explained the task to the observer. After the 2-minute adaptation period, the observer was asked to observe the colour appearance of the samples in the two booths, rating the colour difference of each colour-sample pair using the scale established with the greyscale card. Each observer made the colour difference evaluations under nine pairs of light sources, including a practice trial, a null condition trial that had the reference light source in both booths, six trials that had a test light source in the right booth, and a repeated trial that had a test light source in the right booth for evaluating intra-observer variation. After the initial practice trial, the light source pairs were presented in a random order. The entire experiment took 45 minutes for each observer. In total, there were 1008 evaluations (i.e., 14 observers × 9 pairs of illuminants × 8 pairs of samples).

3. Results and discussion

3.1 Inter- and intra-observer variations

Both intra- and inter-observer variations were characterized using the standardized residual sum of squares (STRESS), with the intra-observer variation comparing the evaluations made by each observer under the repeated trials and the inter-observer variation comparing all the evaluations made by each observer and the mean evaluations (i.e., an average observer). The lower the STRESS value, the smaller the variation. The STRESS values for the intra-observer variation ranged between 9.61 and 26.10, with a mean of 18.10; the STRESS values for the inter-observer variation ranged between 9.64 and 26.31, with a mean of 16.48. The range and the mean STRESS values for both intra- and inter-observer variations were much smaller than past studies evaluating colour differences,^5,6,22 suggesting that the experimental results were highly reliable.

3.2 Perceived colour fidelity

The perceived colour fidelity of each light source was characterized using the average rated colour difference of the eight samples from all the observers, as shown in Figure 4, with a higher rating indicating a higher perceived colour fidelity (note: the rated colour difference values were not rescaled and subtracted from 100, as is typically done for computing colour fidelity). Figure 5 shows perceived colour fidelity versus the calculated colour fidelity for the six test light sources, using both standard ((a) and (b)) and custom ((c) and (d)) colour fidelity calculations.

Figure 4

Average perceived colour difference, together with the 95% confidence interval, of the eight samples under each light source in comparison to the reference SPD, with a greater value for a smaller colour difference.

Figure 5

Perceived colour fidelity vs. calculated colour fidelity for each pair of light sources. (a) R_a; (b) R_f; (c) R_{a_Experiment} and (d) R_{f_Experiment}.

All three pairs of SPDs had smaller differences in both R_f and R_{f_experiment}, and larger differences in both R_a and R_{a_experiment}. The experimental results shown in Figure 4, however, illustrate that only the pair of Red_1 and Red_2 had a significant difference in perceived colour fidelity. This pair had the largest differences in R_f and R_{f_experiment}, supporting IES R_f as a more accurate measure of colour fidelity. When spectral uniformity of the test colour samples was eliminated as a factor by replacing the standardized samples with those used in the experiment, the performance of R_{a_experiment} was still much worse than R_{f_experiment}. This implicates the non-uniformity of CIE U*V*W* as an influential limitation of CIE R_a, corroborating other research.^5,6,16,19

Of particular note is the pair of Green_1 and Green_2. Like the blue pair, it showed no significant difference in perceived colour fidelity, but the correlation between perceived and calculated colour fidelity was actually slightly negative in all cases. This pair had the smallest difference for colour fidelity calculations using the IES R_f framework, and the largest difference in colour fidelity calculations using the CIE R_a framework.

To translate these results into practical advice, differences of up to 10 points for CIE R_a resulted in no perceived difference in colour fidelity, whereas the equivalent value for IES Rf was approximately 3 points. It was not the intent of this work to develop these thresholds, however, so they should not be considered representative of results for a broader range of sources.

3.3 Individual sample colour differences

The correlation between calculated colour differences and perceived colour differences for individual colour samples was also examined, using both CIE U*V*W* and CAM02-UCS. These two colour spaces (and associated chromatic adaptation transformations) are those used in CIE R_a and IES R_f calculations, respectively. Both sets of comparisons are shown in Figure 6, with the Pearson correlation coefficients summarized in Figure 7. The more negative the correlation coefficient, the stronger the correlation – the rated values used the convention of a greater value indicating a smaller colour difference. It can be observed that colour difference calculated in CAM02-UCS and CIE U*V*W had a similar average correlation to the perceived colour difference, but there were notable differences for individual colour samples HL2 and HL4.

Figure 6

Perceived colour difference vs. the calculated colour difference for each colour sample in (a) CAM02-UCS and (b) CIE U*V*W*.

Figure 7

Pearson correlation coefficient between the perceived and calculated colour differences for each sample in CAM02-UCS and CIE U*V*W.

CAM02-UCS had a much higher correlation than CIE U*V*W* for HL2, which is nominally blue (having a high reflectance between 430 and 480 nm). Such a better performance of CAM02-UCS for samples with a high reflectance in this range was never experimentally investigated before, but it is critically important to typical white-light LEDs.

Windisch et al.²¹ noted that R_f is sensitive to small changes in the short-wavelength region, whereas CIE R_a is not. This experiment confirms that R_f is the more accurate representation. The use of CIE R_a allows for more flexibility in binning and is more lenient as an InGaN chip's emission changes with temperature, but it masks real changes in colour fidelity. (The outcome of these changes on subjective evaluations is a separate question, which requires consideration of measures beyond colour fidelity.)

3.4 Uniformity of colour spaces

As shown in Figure 8, the perceived colour difference between Red_1 and Red_2 was mainly due to the individual differences for colour samples HL7 and HL8 (nominally red and purple, respectively). Although the perceived colour difference of HL7 under Red_1 compared to the reference was similar to the other colour samples, the calculated colour difference of HL7 in CIE U*V*W* was much larger, as shown in Figure 9(a). In addition, though the colour differences of HL7 and HL8 under Red_2 versus the reference were rated larger than those under Red_1, the calculated colour differences of these two samples seemed to be overestimated in CIE U*V*W*, as shown in Figure 9(b). It can be concluded that CIE U*V*W* colour space generally enlarged the colour difference along red–green axis (i.e., a′ axis in CAM02-UCS and U* axis in CIE U*V*W*), which caused a larger colour difference in comparison to CAM02-UCS. This discrepancy between CAM02-UCS and CIE U*V*W* in characterizing the colour difference matches has been identified theoretically by others.^16,19 It may be greater at lower CCTs and less at higher CCTs than the nominally 3500 K examined here.¹⁶

Figure 8

Perceived colour difference, together with the 95% confidence interval, for each sample between Red_1 and the reference SPD and between Red_2 and the reference SPD.

Figure 9

Calculated colour difference for each sample under (a) Red_1 and (b) Red_2 in comparison to the reference SPD along the red-green axis in CAM02-UCS and CIE U*V*W*.

For the six test light sources in this study, the average magnitude of the colour shift in comparison to the reference illuminant was 3.50 units in CAM02-UCS, with 0.827, 2.768 and 1.492 units in the J′, a′ and b′ directions, respectively. In contrast, the average magnitude was 2.23 units in CIE U*V*W*, with 0.828, 7.036 and 2.032 units in the W*, U* and V* directions. In other words, CIE U*V*W* colour space is more sensitive to a colour shift along the red-green direction but less sensitive to a colour shift in the yellow-blue direction, in comparison to CAM02-UCS; this explains the stronger effect of red InGaAlP chips on CRI R_a than on IES R_f reported by Windisch et al.²¹ This is also consistent with current recommendations regarding CIE R₉ (saturated red), for which much lower values (larger colour differences in CIE U*V*W*) are considered acceptable.

4. Conclusion

Three pairs of light sources were carefully designed to have different SPDs, similar IES R_f values but very different CIE R_a values. The similarity of average perceived colour differences for each pair of light sources suggested better performance of IES R_f with respect to human perception. Furthermore and in particular, the pair that had a small peak wavelength difference in the nominally red wavelength region revealed that CIE U*V*W*, used in calculating CIE R_a, overestimated the colour shifts versus perception along the red-green axis, which is particularly important to subjective qualities of illumination. In short, the findings reported in this paper suggested a higher robustness of IES R_f in evaluating white LED colour fidelity.

Footnotes

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The Public Policy Research Funding Scheme from Policy Innovation and Co-ordination Office of the Hong Kong Special Administrative Region Government (Project Number: 2017.A6.091.18A). The contribution of Michael Royer was supported by the U.S. Department of Energy Solid-State Lighting Program.

References

Commission International de l'Eclairage. CIE 13.2-1995. Method of Measuring and Specifying Colour Rendering Properties of Light Sources, Vienna: CIE, 1995.

Jost-Boissard S, Fontoynont M, Blanc-Gonnet J. Colour rendering of LED sources: Visual experiment on difference, fidelity and preference: CIE Light and Lighting Conference with Special Emphasis on LEDs and Solid State Lighting. Budapest, Hungary, 27–29 May, 2009: pp. 55–62.

Royer

Wilkerson

Wei

Houser

Davis

. Human perceptions of colour rendition vary with average fidelity, average gamut, and gamut shape. Lighting Research and Technology 2016; 49: 966–991.

Royer M, Wilkerson A, Wei M. Human perceptions of colour rendition at different chromaticities. Lighting Research and Technology 2018; 50(7): 965–994.

Wei

Smet

KAG

Lin

. The prediction of perceived colour differences by colour fidelity metrics. Lighting Research and Technology 2017; 49: 805–817.

Jost

Cauwerts

Avouac

. CIE 2017 colour fidelity index Rf: a better index to predict perceived colour difference?. Journal of the Optical Society of America A 2018; 35: B202–B213.

Zhang

Feng

. Toward a unified model for predicting colour quality of light sources. Applied Optics 2017; 56: 8186–8195.

Smet

Hanselaer

. Memory and preferred colours and the colour rendition of white light sources. Lighting Research and Technology 2016; 48: 393–411.

Wei

Houser

Allen

Beers

. Colour preference under LEDs with diminished yellow emission. Leukos 2014; 10: 119–131.

10.

Wei

Houser

David

Krames

. Colour gamut size and shape influence colour preference. Lighting Research and Technology 2017; 49: 992–1014.

11.

Wei

Houser

. Systematic changes in gamut size affect color preference. Leukos 2017; 13: 23–32.

12.

Lin

Wei

Smet

KAG

Tsukitani

Bodrogi

Khanh

. Colour preference varies with lighting application. Lighting Research and Technology 2017; 49: 316–328.

13.

Commission Internationale de l'Eclairage. CIE 224:2017. CIE 2017 Colour Fidelity Index for Accurate Scientific Use, Vienna: CIE, 2017.

14.

Illuminating Engineering Society of North America. IES TM-30-15. IES Method for Evaluating Light Source Color Rendition, New York, NY: IESNA, 2015.

15.

David

Fini

Houser

Ohno

Royer

Smet

KAG

Wei

Whitehead

. Development of the IES method for evaluating the colour rendition of light sources. Optics Express 2015; 23: 15888–15906.

16.

Smet

AKG

David

Whitehead

. Why colour space uniformity and sample set spectral uniformity are essential for colour rendering measures. Leukos 2016; 12: 39–50.

17.

Commission Internationale de l'Eclairage. CIE 224: 2017. CIE 2017 Colour Fidelity Index for Accurate Scientific Use, Vienna: CIE, 2017.

18.

Illuminating Engineering Society of North America. IES TM-30-18. IES Method for Evaluating Light Source Color Rendition, New York, NY: IESNA, 2018.

19.

Royer

. Comparing measures of average color fidelity. Leukos 2017; 14: 69–85.

20.

Rea M, Freyssinier JP. TM-30: Curb Your Enthusiasm. Lighting Design and Application. New York, NY: Illuminating Engineering Society, 2015: pp. 11–12.

21.

Windisch

Heidel

Binder

Bergenek

. Impact of spectral features of common LED lighting systems on TM-30 colour indices. Optics Express 2017; 25: 1824–1830.

22.

Luo

Liu

. Testing different colour rendering metrics using colour difference data. Lighting Research and Technology 2017; 49: 539–560.

23.

Esposito

Houser

. Models of colour quality over a wide range of spectral power distributions. Lighting Research and Technology. Epub ahead of print 13 April 2018. DOI: 1477153518765953.

24.

Wei

Houser

David

Krames

. Colour gamut size and shape influence colour preference. Lighting Research and Technology 2017; 49: 992–1014.

25.

Guan

Luo

. Investigation of parametric effects using small colour differences. Color Research and Application 1999; 24: 331–343.

26.

Luo

Rigg

. Chromaticity-discrimination ellipses for surface colours. Color Research and Application 1986; 11: 25–42.

27.

Guan

Luo

. A colour-difference formula for assessing large colour differences. Color Research and Application 1999; 24: 344–355.