Colour gamut size and shape influence colour preference

Abstract

Two psychophysical experiments were conducted to investigate the effect of colour gamut size and shape on colour preference. In the first, two side-by-side booths were designed to resemble a retail setting with clothing; in the second, a single booth was designed to resemble a restaurant setting, but also included a mirror to permit the observer's evaluation of skin tone. These settings were illuminated with two sets of sources, where, compared to a fixed reference, one set created modest chroma enhancement and the other set created greater increases in object chroma. Within each set, gamut shape varied, meaning different hues were saturated, even though, on average, the spectra created the same average increase in chroma. When objects were unfamiliar, as with the fabrics, all chroma-enhancing spectra were preferred to the fixed reference regardless of the gamut shapes. When familiar objects were present, such as food, observers were more discerning about changes in chroma and hue. We conclude that a graphic of gamut shape is an important adjunct to average measures of colour fidelity and gamut.

1. Introduction

Since 1965, the CIE General Colour Rendering Index R_a¹ has been used by lighting manufacturers to guide the spectral design of light sources, and by specifiers to gauge the suitability of sources for design applications. The shortcomings of R_a are well known, especially for sources whose spectral power distributions (SPDs) have sharp peaks and valleys, which includes some LEDs.^2–5

The past decade has seen numerous efforts to better characterize colour rendition of light sources.^6,7 Most agree that a measure of colour fidelity is not enough if used alone, as higher fidelity is not always preferred.^8–15 People often prefer illumination that enhances object chroma.^8,9,12,15,16 The colour coordinates of a set of test-colour samples within a colour space can be used to compute a colour gamut area, which can be employed as a proxy of the average change of chroma under a given source, in comparison to a reference illuminant.⁶ Thus, several two-measure systems, comprised of a fidelity measure and a gamut measure, have been proposed.^6,10,17–21

Recently, a new method was developed by the Colour Metrics Task Group in the Illuminating Engineering Society of North America (IES), leading to the publication of TM-30-15.^17,18,22 This method includes a fidelity index (R_f) and a relative gamut index (R_g). R_f is designed to be a more accurate version of R_a, characterizing how similar 99 colour evaluation samples (CES) appear under a test source and a reference illuminant; R_g estimates the average change of chroma of the 99 CESs under a test source in comparison to a reference illuminant (for details, see^17,18).

Two sources having similar R_f and R_g values, however, can render colours differently. The enhanced chroma in certain hues can be offset by a decrease of chroma in other hues, resulting in a similar gamut area (i.e. R_g) but different gamut shapes.²³ Such a difference may be critical when the enhancement happens in some specific hues, as suggested by past studies,^8,9,12 but it cannot be conveyed by R_f and R_g values. Thus, the IES TM-30-15 method also includes colour vector and distortion graphics, which visually illustrate hue and chroma shifts.^17,18

In this study, we hypothesized that sources with similar measures of relative gamut, but different gamut shapes would lead to significantly different colour preferences. Two sets of spectra were designed and evaluated in viewing booths that were designed with objects characteristic of restaurant and retail settings. To our knowledge, the effect of gamut shape has not been previously investigated or reported.

2. Method

2.1. SPDs

A reference spectrum and two sets of enhanced spectra were designed, as shown in Figure 1. All SPDs were metameric to a 3000 K blackbody radiator using the CIE 1964 10 ° colour matching functions (CMFs).²⁴ We found that this yielded a fairly good perceptual match of chromaticity, whereas using the CIE 1931 2 ° CMFs gave unacceptable results in terms of chromaticity matching. In each series, the SPDs had a similar relative gamut score, but different gamut shapes.

The relative gamut score was calculated with the computational engine from the colour quality scale (CQS),¹² modified to use a sample set of 370 real objects that were selected from an existing dataset²⁵ and that are uniformly distributed in the CAM02-UCS.²⁶ This sample set was created during the development of IES TM-30-15.^17,18 The method used to compute fidelity and relative gamut scores reflects what we believed to be the best practice at the time this experiment was conceptualized. The colorimetric characteristics of the spectra are summarized in Table 1. Each SPD in the 110-series has a relative gamut score of approximately 110; each in the 120-series has a relative gamut score of approximately 120. TM-30-15 was published after this experiment, prompting us to perform post hoc computations for IES R_f and R_g, which are also included in Table 1. The values of R_f and R_g show some variations within a series. Gamut area index (GAI), which is an absolute measure of gamut based on comparison to an equal-energy reference, was also investigated.

Table 1

Summary of the colorimetric characteristics of the spectra included in the study

	1931 CMFs		CCT	D _uv	CRI		GAI	Based on the 370 real object samples		IES TM-30-15
	x	y	CCT	D _uv	R _a	R ₉	GAI	Fidelity	Relative gamut	R _f	R _g
Reference	0.4189	0.3745	3072	−0.0098	86	57	72	87	101	88	97
120-A	0.4180	0.3864	3197	−0.0044	79	82	86	67	121	63	117
120-B	0.4269	0.3857	3021	−0.0061	73	76	85	72	120	66	118
120-C	0.4214	0.3823	3093	−0.0067	65	27	89	73	121	70	117
120-D	0.4215	0.3720	2997	−0.0113	53	−28	95	70	122	68	116
120-E	0.4263	0.3646	2838	−0.0151	37	−91	97	57	122	59	113
110-A	0.4163	0.3841	3211	−0.0051	88	71	79	84	111	81	108
110-B	0.4171	0.3824	3180	−0.0060	87	92	80	85	111	84	107
110-C	0.4153	0.3758	3156	−0.0086	82	75	84	85	110	84	108
110-D	0.4193	0.3754	3072	−0.0094	78	55	83	83	111	83	106
110-E	0.4213	0.3720	3003	−0.0112	72	23	84	80	111	79	105
110-F	0.4206	0.3698	2995	−0.0121	64	−12	85	72	109	72	104

Note: the spectra were designed to have chromaticities close to a 3000 K blackbody radiator when computed with the CIE 1964 10 ° colour matching functions.

The chroma in red hues increases as the 110 series goes from “A” to “F” and as the 120 series goes from “A” to “E”, which can be observed using the IES TM-30-15 colour vector graphics. The gamut shapes for the 120-series are shown in Figure 2; the 110-series is shown in Figure 3, and the reference spectrum in Figure 4. All of these enhanced spectra introduced average chroma enhancement in comparison to the reference, with relative gamut scores greater than 100.

All SPDs were realized using a spectral-tunable light source (Telelumen Light Replicator) and measured with a diffuse reflectance standard (Labsphere SRT-MS-100, ρ = 99%) that was placed in the viewing booths described below.

2.2. Apparatus

This study comprised two experiments. The first employed side-by-side booths designed to resemble a retail setting, as illustrated in Figure 5. The second, which occurred four months later, employed a single booth designed to resemble a restaurant setting, as shown in Figure 6.

Each booth in the retail-setting experiment had dimensions of 53 cm (width) × 38 cm (depth) × 81 cm (height). The bottom of the front side was open, with the top portion covered to prevent observers from seeing the source of the illumination. A TeleLumen Light Replicator was placed above each booth, providing uniform illumination through the circular aperture at the centre of the ceiling of the booth. The two booths had similar luminance distribution, as verified with calibrated high dynamic range (HDR) imaging. The booth interiors were painted with Munsell N8 paint. The floor of each booth was covered by six shirts in different colours and the back wall was covered by seven ties in different colours, as shown in Figure 5. The contents were selected to be representative of objects found in a generic retail environment, such as a department store. A chin and forehead rest was mounted outside the two booths (43.2 cm away from the front opening of the booth), aligning the sagittal plane of an observer with the panel dividing the two booths. The rest was 17.8 cm above the floor of the booth, allowing observer to easily see all of the objects.

The single booth in the restaurant setting experiment had dimensions of 81 cm (width) × 41 cm (depth) × 104 cm (height). The interior of the booth was painted with white matte paint (Behr Premium Plus Ultra Paint and Primer in One), which had a relatively flat reflectance distribution across the visible spectrum. A TeleLumen Light Replicator was placed above the booth, providing uniform illumination through a circular aperture at the centre of the ceiling panel of the booth. The front side of the booth was partially covered, with the bottom 61 cm open, preventing observers from seeing the TeleLumen directly. A chin–forehead rest was mounted on the floor of the booth, centred on the opening. The floor of the booth was partially cut, allowing the observer to be immersed in the light setting and to increase the light level on the face of the observer.

As shown in Figure 6, various objects were placed in the booth, including a cutting board, cured meats, bread, olives (placed in a bowl), sliced peppers (placed in a bowl), flowers, sushi (placed on a plate), red salt shaker, drink bottle with a label, napkin, and chopsticks. A mirror was also placed in the booth, leaning against the back wall, allowing the observers to evaluate the skin tone of their face. The arrangement of the booth was designed to resemble a restaurant setting, where both facial and object rendition are important.

2.3. Observers

Forty observers (30 males and 10 females) and 40 other observers (23 males and 17 females) were recruited for the retail and the restaurant settings, respectively. Observers were between 19 and 25 years (mean = 21.5 years, standard deviation = 1.80 years) and had normal colour vision as tested by the 24 Plate Ishihara Colour Vision Test. For the retail setting, 30 observers self-identified as Caucasian, 8 Asian, 1 Hispanic/Latino, and 1 mixed; for the restaurant setting, 26 self-identified as Caucasian, 12 Asian and 2 Black.

2.4. Experimental design

2.4.1. Retail setting

The independent variable was the SPD of the light setting. As shown in Appendix A, the dependent variables were 10 colour preference ratings between the two booths using a six-point rating scale, which did not have a neutral point as a precaution against a potential contraction bias.^27,28 Each observer made evaluations under 13 pairs of light settings, which were divided into two sets. The first set comprised six pairs, with five pairs between the reference spectrum and each 120-series spectrum and a null condition pair (i.e., the reference spectrum in both booths); the second set comprised seven pairs, with six pairs between the reference spectrum and each 110-series spectrum and a null condition pair. The order of the pairs within each set was randomized. The order of the two sets was counterbalanced between observers. The location of the reference spectrum and the enhanced spectra was counterbalanced between experimental days. All observers on the same day observed the same left/right presentations. The horizontal illuminance at the floor of the centre of each booth was calibrated at 300 ± 10 lux. The standard error of the luminance values for the floor of the booth was 0.005, which was derived from a HDR photograph.

2.4.2. Restaurant setting

The independent variable was the SPD of the light settings. The dependent variable was a colour preference evaluation between a pair of light settings. Pairs of light settings were presented in a sequential mode, with each stimulus being presented for 5 s with a dark period of 0.01 s between the two stimuli. The observer was asked to choose which of the two stimuli in each pair he or she preferred, with reference to the overall colour appearance of the objects in the booth, and separately, his or her skin tone as observed in the mirror. The responses were forced choices.

Only five SPDs were used in each enhanced series. This was because the results from the retail-setting experiment indicated a relatively small difference between SPDs 110-A and 110-B. Since an all-possible-pairs method was employed in the restaurant setting session, 110-B was dropped to keep the length of the experiment reasonable for observers. All-possible-pairs of SPDs were presented to each observer, including 10 pairs of mixed stimuli (i.e., two different spectra) and 5 pairs of null condition trials (i.e. two identical spectra). In order to counter a possible interval bias or order bias, the 10 pairs of mixed stimuli were presented in two orders (e.g. AB and BA), which was counterbalanced between observers.

In total, each participant made 60 forced choice evaluations. Thirty evaluations were judgments of skin tone (15 under each pair in the 120-series and 15 under each pairs in the 110-series). The other 30 were judgments of objects. The order of skin tone evaluation and objects evaluation were counterbalanced between observers. The horizontal illuminance at the floor of the centre of the booth was calibrated at 500 ± 20 lux; the vertical illuminance at the plane of an observer's face was approximately 750 lux. The standard error of the luminance values for the floor of the booth was 0.016, which was derived from an HDR photograph.

2.5. Procedure

2.5.1. Retail setting

Upon arrival, the observer read a brief description of the experiment, signed an informed consent form, and completed a general information survey (e.g. sex, age, and race) under illumination from 3000 K linear fluorescent lamps at about 300 lux. The participant was then escorted into the experiment space and seated in front of the side-by-side booths; both booths were under illumination of the reference spectrum. The room lighting was switched off, such that the lighting within the booths was the only illumination during the experimental trials. He or she was instructed to adjust the height of the seat and to comfortably place his or her head in the chin and forehead rest. The experimenter read the instructions from a pre-written script and answered questions raised by the participant.

Before evaluating all the pairs in each set, the Telelumen for one booth was programed to present each of the 12 spectra (one reference spectrum, five 120-series spectra, and six 110-series spectra) for 10 s in random order; the entire cycle was repeated twice. During this period, the Telelumen for the other booth was set to present the reference spectrum and the observer was asked to compare the appearance of all the objects between the two booths, allowing the observer to have an idea about the range of the conditions to be experienced.

After presenting all possible conditions to the participant, the experimenter set the light settings according to a pre-written script. The observer was asked to carefully compare the appearance of the objects between the two booths for 30 s, allowing the visual system to adapt to the condition. Then the questionnaire shown in Appendix A was given to the observer. After answering all questions, the experimenter took the questionnaire back and proceeded to the next pair of light settings. The same procedure was followed for all comparisons. After finishing this first set, the observer was asked to leave the seat and experience a three-minute wash-out period under the illumination of 3000 K linear fluorescent lamps at 300 lx.

After the wash-out period, the participant returned to the booth and finished the other set of the comparisons. The same procedure described above was followed.

After finishing all comparisons, the questionnaire shown in Appendix B was given to the observer to finish. Then, he or she was escorted to a table where the 24 Plate Ishihara Colour Vision test was administered. The entire procedure for the two sessions took 15 to 20 min for each observer.

2.5.2. Restaurant setting

Upon arrival, the observer read a brief description of the experiment, signed an informed consent form, and completed a general information survey (e.g. sex, age, and race) under the illumination of 3000 K linear fluorescent lamps at about 300 lux. The observer was then escorted and seated in front of the booth, with his or her chin and forehead comfortably fixed on the rest. Except for the lighting of the booth, there was no illumination in the space. The experimenter read the instructions from a script and answered questions raised by the observer. The experimenter instructed the observer that the following evaluations should be made based on the colour appearance of the objects in the booth or the skin tone of his or her face seen in the mirror. The observer was asked to not focus on the skin tone seen in the mirror when evaluating the colour appearance of the objects, and to not focus on the appearance of the objects when evaluating the skin tone. Half of the observers completed skin tone evaluations first, the other half completed object evaluations first.

The experiment proceeded with three practice pairs, which covered different conditions and were identical for all the observers. The observer was free to ask any questions. After answering the questions, the experiment continued with the presentation of the 30 pairs listed on the recording sheet, with 15 pairs of 120-series and 15 pairs of 110-series. The order of the 120-series and the 110-series was counterbalanced between participants (i.e. 20 participants evaluated the 15 pairs of 120-series first and 20 participants evaluated the 15 pairs of 110-series first). The order of the 15 pairs within the 120-series and 110-series was randomized.

For each pair of stimuli, the experimenter loaded a software script that alternated the lighting in the booth every 5 s with a dark period of 0.01 s between each alternation. The dark period was included to let the participant know the lighting was changing, especially for null condition pairs when the two stimuli were identical. As the lighting was changing, the experimenter spoke aloud ‘A, B, A, B…’. The observer was instructed to observe at least four alternations before providing a response, but he or she was free to observe and compare the two stimuli within each pair as long as necessary. After the observer made a selection, the experimenter recorded the selection and proceeded to the next pair. After completing the 30 paired comparisons, the observer finished the two questionnaires shown in Appendix C. The observer finished the questionnaires sequentially and was unaware that a second questionnaire would be coming. The observer then left the booth to experience a three-minute wash-out period.

After the wash-out period, the observer returned to the booth and finished the other 30 pairs of evaluations. If the observer had evaluated his or her skin tone first, the second set was for evaluation of the objects in the booth, and vice-versa.

After finishing all the evaluations, the observer stepped out of the space and performed another 15-min experiment that is not described here. The 24 Plate Ishihara Colour Vision Test was then administered in a windowless room. The entire procedure took about 75 min for each observer. The food was purchased every other day from the same grocery store, with an effort to minimize the variation. The flowers were purchased every three days and did not appreciably change during the six days of data collection.

3. Results

3.1. Retail setting

The rating scales that appeared in the questionnaires were converted to numerical scales for statistical analyses: 1. Strongly prefer the reference spectrum; 2. Moderately prefer the reference spectrum; 3. Slightly prefer the reference spectrum; 4. Slightly prefer the enhanced spectrum; 5. Moderately prefer the enhanced spectrum; 6. Strongly prefer the enhanced spectrum.

No significant difference was observed between the two booths for the null condition trials, when both booths were under the illumination of the reference spectrum. This was tested using a one-sample t-test (p =0.160 for the 120-series; p = 0.643 for the 110-series).

3.1.1. Enhanced spectra in the 120-series

The mean ratings of the preference evaluations for different colours between the reference spectrum and each enhanced spectrum in the 120-series are shown in Figure 7; one-sample t-tests were employed to test if the mean ratings were significantly different from 3.5. Overall preference, preference of red, orange, green, cyan, blue, pink under each 120-series spectrum were significantly higher than 3.5, indicating a higher preference for colours under these 120-series spectra. For yellow, only 120-B and 120-C were preferred. For skin tone, 120-A, 120-B, 120-C, and 120-D were preferred.

A repeated-measures ANOVA was employed to test the effect of SPD on the evaluation for each colour, as summarized in Table 2 (when the assumption of sphericity was violated, the Huynh-Feldt or Greenhouse-Geisser correction was employed). For the colours where SPD was a significant factor, pair-wise comparison with a Bonferroni adjustment was employed (family-wise error rate was set at 5%), from which:

Red: 120-E was rated significantly higher (more preferred) than 120-A;

Orange: 120-D and 120-E were rated significantly higher (more preferred) than 120-A;

Cyan: 120-E was rated significantly higher (more preferred) than 120-A and 120-B.

Table 2

Summary of statistical analyses regarding the effect of SPD in the 120-series on colour preference evaluations

	Mauchly's test of sphericity		Effect of SPD
Overall	χ² (9) = 20.58;	p = 0.015	F (3.23) = 2.30;	p = 0.076
Red	χ² (9) = 14.11;	p = 0.119	F (4) = 7.00;	p < 0.001
Orange	χ² (9) = 17.09;	p = 0.048	F (3.37) = 5.65;	p = 0.001
Yellow	χ² (9) = 12.65;	p = 0.180	F (4) = 1.45;	p = 0.220
Green	χ² (9) = 9.11;	p = 0.428	F (4) = 1.26;	p = 0.287
Cyan	χ² (9) = 16.61;	p = 0.055	F (4) = 5.36;	p < 0.001
Blue	χ² (9) = 12.11;	p = 0.207	F (4) = 0.33;	p = 0.857
Pink	χ² (9) = 44.92;	p < 0.001	F (2.79) = 1.80;	p = 0.156
Skin	χ² (9) = 36.43;	p < 0.001	F (2.94) = 1.49;	p = 0.247

3.1.2. Enhanced spectra in the 110-series

The mean ratings of the preference evaluations for different colours between the reference spectrum and each enhanced spectrum in the 110-series are shown in Figure 8; one-sample t-tests were employed to test if the mean ratings were significantly different from 3.5. Overall preference, and preference for red, orange, green, cyan, blue, and pink were significantly higher than 3.5, indicating a higher preference for these colours under the 110-series spectra. For yellow, only 110-A, 110-B, and 110-E were preferred. For skin tone, 110-E and 110-F were preferred.

Similar repeated-measures ANOVA, together with pair-wise comparisons, were conducted on the evaluations, as summarized in Table 3, from which we found:

Red: 110-E and 110-F were rated significantly higher (more preferred) than 110-A; 110-E was rated significantly higher than 110-B;

Orange: 110-F was rated significantly higher (more preferred) than 110-A, 110-B, and 110-C.

Table 3

Summary of statistical analyses regarding the effect of SPD in 110-series on colour preference evaluations

	Mauchly's test of sphericity		Effect of SPD
Overall	χ² (14) = 26.38;	p = 0.023	F (3.77) = 0.83;	p = 0.503
Red	χ² (14) = 24.83;	p = 0.037	F (4.03) = 5.68;	p < 0.001
Orange	χ² (14) = 18.37;	p = 0.191	F (5) = 7.63;	p < 0.001
Yellow	χ² (14) = 31.06;	p = 0.006	F (3.83) = 1.47;	p = 0.217
Green	χ² (14) = 11.09;	p = 0.680	F (5) = 1.33;	p = 0.254
Cyan	χ² (14) = 14.81;	p = 0.392	F (5) = 1.35;	p = 0.245
Blue	χ² (14) = 14.07;	p = 0.445	F (5) = 0.82;	p = 0.540
Pink	χ² (14) = 18.98;	p = 0.166	F (5) = 1.80;	p = 0.114
Skin	χ² (14) = 42.57;	p < 0.001	F (3.16) = 1.28;	p = 0.284

3.1.3. Purchasing experience

A questionnaire related to shopping experience was given to the observer after all pairs of light settings had been evaluated (Appendix B). Figure 9 summarizes mean ratings and 95% confidence intervals for the three questions asked. The results suggest that the range of light spectra evaluated was significant enough to influence where a person did their shopping, the pleasantness of the shopping experience, and recommendations about where to shop.

3.2. Restaurant setting

This session was designed with counterbalancing, where the SPDs included in each pair had an even chance to be presented first or second. Random chance suggests an even number of selections for first/second throughout the experiment. The first stimulus in the pair was selected 49.9% of the time, which was not statistically different from 50% (p = 0.92).

3.2.1. Preference for food and objects

The effects of SPD on preference were examined with variance stable rank sums (VSRS), an adaptation of two-way analysis of variance by ranks developed by Dunn-Rankin et al.²⁹ and used in past lighting research.^24,30–32 Table 4 summarizes the results of VSRS tests for the evaluations of food, flowers, and other objects. For the 120-series, observers preferred colours under 120-B, 120-C, and 120-D over those under 120-E, and those under 120-C over those under 120-A; for the 110-series, the observers preferred the colours under 110-D, 110-E, and 110-F over those under 110-A.

Table 4

Summary of the VSRS tests for the food, flowers, and objects in the restaurant setting

120-series		120-E	120-A	120-B	120-D	120-C
	Ri	48	68	88	90	106
120-E	48
120-A	68	20
120-B	88	40	20
120-D	90	42	22	2
120-C	106	58	38	18	16

110-series		110-A	110-C	110-D	110-F	110-E
	Ri	50	80	84	90	96
110-A	50
110-C	80	30
110-D	84	34	4
110-F	90	40	10	6
110-E	96	46	16	12	6

Note: Ri is the total number of selections with a maximum value in this experiment of 160 (4 comparisons between each condition and every other condition and 40 observers: 4 × 40 = 160). An Ri value of 160 would indicate that that lighting condition had always been selected over every other condition. Values in the matrix are the difference of Ri between each pair of SPDs. Shaded values represent comparisons that are significantly different at the α = 0.05 level (with a value larger than 31.10). When the value is larger than 37.11, it is significant at the α = 0.01 level.

3.2.2. Preference for skin tone

Similarly, VSRS was employed to test the effect of SPD on skin tone preference. As summarized in Table 5, for the 120-series, 120-C, 120-B, 120-A, and 120-D were preferred over 120-E; 120-B and 120-C were preferred over 120-D; and 120-B and 120-C were preferred over 120-A. For the 110-series, 110-A, 110-C, and 110-D were preferred over 110-E; 110-C and 110-D were preferred over 110-F.

Table 5

Summary of the VSRS tests for skin tone preference.

120-series		120-E	120-D	120-A	120-B	120-C
	Ri	13	61	82	120	124
120-E	13
120-D	61	48
120-A	82	69	21
120-B	120	107	59	38
120-C	124	111	63	42	4

110-series		110-E	110-F	110-A	110-D	110-C
	Ri	51	56	86	93	114
110-E	51
110-F	56	5
110-A	86	35	30
110-D	93	42	37	7
110-C	114	63	58	28	21

Note: Ri is the total number of selections with a maximum value in this experiment of 160 (4 comparisons between each condition and every other condition and 40 participants: 4 × 40 = 160). An Ri value of 160 would indicate that that lighting condition had always been selected over every other condition. Values in the matrix are the difference of Ri between each pair of SPDs. Shaded values represent comparisons that are significantly different at the α = 0.05 level (with a value larger than 31.10). When the value is larger than 37.11, it is significant at the α = 0.01 level.

3.2.3. Rating and rank of influential objects

After completing all pairs of comparisons while observing the array of objects, each observer was asked to complete two questionnaires (Appendix C). The first questionnaire asked the observer to reflect upon how each object contributed to his or her overall judgment. The second questionnaire asked the observer to order the objects from most to least important in contributing to his or her judgment.

Mean ratings and rankings, together with 95% confidence intervals, are summarized in Figures 10 and 11, respectively. The peppers and the sushi most strongly affected evaluations, followed by the cured meats and flowers. The mean ratings and the mean ranks of these objects were highly correlated (Pearson correlation r = −0.979, p < 0.001). The correlation has a negative sign because the scales had opposite polarity.

4. Discussion

4.1. Preference for enhanced chroma

All enhanced spectra included in this study had relative gamut scores greater than 100, indicating an average ability to enhance chroma, in comparison to the reference spectrum. In the retail setting, overall colour preference was rated higher under each enhanced spectrum, and most colours were preferred under these enhanced spectra, regardless of gamut shape. This preference for enhanced chroma is consistent with past studies.^8,9,12 Though oversaturation may reduce preference,³³ our results suggest that the spectra included in this study did not create too much saturation.

4.2. Effect of gamut shape

The idea that gamut shape would affect colour preference was not supported by the results from the retail setting. Though different chroma enhancements occurred under each of the enhanced spectra, as shown in Figure 12, these did not result in different ratings of preference. There are two likely explanations for this result. First, the comparisons were always between the reference and one of the enhanced spectra, which was an intentionally conservative methodology. Participants never made comparisons between pairs of enhanced spectra. Had we forced participants to choose between different pairs of enhanced spectra, it is more likely that differences would have been found between them. This method was also selected for practical considerations. The 3000 K reference produced colour rendition that might be considered typical of common sources in use today (CIE R_a = 86, IES R_f = 88, IES R_g = 97). We were especially interested in whether a gamut-enhancing spectra would be preferred, in comparison to a typical reference. Second, the shirts and ties are dyed fabrics with man-made colours that were unfamiliar to the observers. With no obvious anchors or memories to suggest a “true colour” for these fabrics, the observers exhibited a general preference for higher chroma. This latter idea motivated us to investigate the colour preference in a restaurant setting, which included more familiar and natural objects, for which observers would have internal anchors and expectations about colour appearance.

In the restaurant setting, all objects showed shifts in chromaticity (Figure 13) but some objects influenced observers' evaluations more than others, as shown in Figures 10 and 11. Objects with warm colours (e.g. red, pink, and orange) generally had more influence than other colours. The importance of chroma enhancement in red colours found here is consistent with other studies.^8,9 Since the spectra in each series had similar gamut scores (Table 1), the significant differences in colour preference are largely due to differences in gamut shape. As can be observed in Figures 2 and 3, the chroma enhancement in blue/yellow colours under 120-A and 110-A was highest in each series, but the preference of these two were lower than those which enhanced chroma in red colours. Not all the red objects were ranked equally important. For example, the red salt shaker was not rated as important as sushi, peppers, and cured meats. Natural objects were generally given higher importance. Compared to the retail settings, the observers were more likely to have internal anchors and expectations about colour appearance for the natural objects presented in the restaurant setting.

The combined importance of chroma enhancement in red colours and the colour appearance of natural objects can at least partially explain the lower preference of 120-E. Higher chroma for red colours is not always preferred, as the colours may be over-saturated and appear unnatural. There seems to have been an optimal colour shift for the objects in the restaurant setting, which was achieved by 120-C, 110-E, and 110-F. As shown in Figure 14, the chromaticity coordinates in CAM02-UCS of peppers and sushi were close to each other under the illumination from these three sources. In comparison to the reference spectrum, 120-C, 110-E, and 110-F introduced similar colour shifts comprising moderate chroma enhancements, and small hue shifts. Therefore, the design of an optimized saturation-enhancing light source may have to rely on the specific colour shifts (direction and magnitude) for specific colours or objects, rather than on average values like R_g.

120-E also had the lowest fidelity score (CIE R_a = 37, IES R_f = 59). While this may have also played a role in why this source was less preferred than other sources, the data also make clear that a fidelity index alone, such as CIE R_a or IES R_f, cannot predict preference. All enhanced spectra in both the 110- and 120-series had CIE R_a and IES R_f values less than that of the reference (Table 1), yet all were preferred to the reference. There are also numerous conditions were the results of the VSRS tests (Table 4) suggest higher preference for a source with a lower fidelity score. For example, 110-C (R_a = 78), 110-D (R_a = 72), and 110-E (R_a =64) were all rated more highly than the reference (R_a = 88).

The preferred spectra for skin tone were not identical to those for the objects. Figure 15 illustrates chromaticities of Asian and Caucasian skin under different spectra, which were calculated using SPD of the light settings and the spectral reflectance distribution of typical Asian and Caucasian skin.³⁵ Among the five spectra in each series, 120-C and 110-C created the smallest hue shift for both Asian and Caucasian skin tone. For skin tone rendition, maintaining minimal hue shifts seemed to be more important than chroma enhancement. This is different than what was found with the natural objects such as salmon, sushi, and peppers, where moderate chroma enhancements were rated as more preferred.

4.3. Characterizing colour preference

This study included three different types of evaluations: fabrics (retail setting), natural objects such as food and flowers (restaurant setting), and skin tone. For fabrics, enhanced chroma was generally preferred regardless of hue shift. For natural objects, moderate chroma enhancement with specific colour shifts was preferred – specifically, this applied to those in warm colours (e.g. red, orange) which drove the preference decision. For skin tone rendition, a minimal hue shift was most desired. These results suggest that observers had different criteria for evaluating colour preference in different applications, which corroborates a recent study.³⁶

The dependence of colour preference on lighting application is an opportunity and a challenge. There is an opportunity for manufacturers to develop application-specific lighting products, which produces an appropriate degree of chroma enhancement for specific applications. Understanding different preferences in different applications may complicate the process of lighting design and specification, which is a challenge. But, the potential benefit is illuminated environments that are more preferred by occupants.

The specific fidelity and gamut measures employed in a two-measure system are a critical consideration. As shown in Table 1, the five SPDs in the 120-seris had similar values for R_g (range: 113 to 118) and relative gamut using the 370 samples (range: 120 to 122), but their GAI²¹ values varied considerably (range: 85 to 97). Though 120-A was the only SPD in the 120-series at the boundary of ‘Class A’ source criteria (i.e. R_a ≥ 80, 80 ≤ GAI ≤ 100),^21,37 it was not the most preferred source in the restaurant setting as shown in Tables 4 and 5. Further, 110-E and 110-F do not meet ‘Class A’ criteria, but they were preferred to over 110-A, which is on the boundary of ‘Class A’ criteria, as shown in Table 4. Source 110-C is the only source that unambiguously meets ‘Class A’ criteria with R_a = 82 and GAI = 84, yet it was not statistically different than any other source considered in the evaluation of food, flowers, and objects in the restaurant setting (Table 4). Sources that are optimized to meet the ‘Class A’ criteria tend to reduce chroma in reds, as the non-uniformity of the CIE U × V × W colour space makes R_a disproportionately penalize chroma enhancement in red.³⁸ Such a flaw of ‘Class A’ criteria contradicts the importance of red colours in colour preference evaluation.

These data also illustrate the importance of a colour vector graphic, such as the ones reported in Figures 2 through 4, which are based on IES TM-30-15. The colour vector graphic depicts hue and chroma shifts around the entire hue circle, succinctly depicting information that cannot be conveyed with simple indices that are based on mean scores. It can be used to visually evaluate colour rendition beyond average values.^22,39

5. Summary and conclusion

The colour preference of fabrics arranged in a retail setting, natural and man-made objects arranged in a restaurant setting, and skin tone were studied under two series of 3000 K enhanced spectra. The spectra in each series had a similar relative gamut score (120-series and 110-series) but different gamut shapes.

In the retail setting, the enhanced spectra were compared to a reference spectrum that produced chroma similar to that of a 3000 K blackbody radiator. The overall colour appearance under all of the enhanced spectra was preferred to those under the reference spectrum; no significant difference was observed between the enhanced spectra in each series.

In the restaurant setting, significant differences were found between the spectra in each series and natural objects (especially those in warm colours) were rated to be more influential on observers' judgements, indicating the effect of gamut shape on colour preference.

The differences found between the retail and restaurant settings suggest the dependence of colour preference on application. The observers had no knowledge about the appearance of the objects in the retail settings and generally preferred the enhanced chroma, while the observers were familiar with the appearance of natural objects in the restaurant setting and skin tone and evaluated the colour preference based on chroma enhancement and hue shift.

The effect of gamut shape on colour preference and the difference between lighting applications reported here are important to LED spectral optimization and characterization of light source colour rendition. It illustrates the importance of the colour vector graphic in evaluating colour rendition of a light source. Though a two-measure system (such as IES R_f and R_g) adds an additional dimension about average change in chroma compared to a single fidelity measure, these two measures still have the limitations of averaging. The colour vector graphic conveys crucial information which cannot be presented by two average values, especially when certain colours are important or dominant for an application.

Footnotes

Declaration of conflicting interests

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

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was supported by Soraa, Inc. under a sponsored research agreement with The Pennsylvania State University.

Appendix A

Appendix B

Appendix C

References

Commission internationale de l'Eclairage. Methods of Measureing and Specifying Colour Rendering Properties of Light Sources, Vienna: CIE, 1965.

Commission Internationale de l'Eclairage. CIE 177:2007 Colour Rendering of White LED Light Sources, Vienna: CIE, 2007.

Houser

Mossman

Smet

KAG

Whitehead

. Color rendering and its applications in lighting. Leukos 2016; 12: 7–26.

Houser

. Editorial: if not CRI, then what? Leukos 2013; 9: 151–153.

Wei

Houser

. Status of solid-state lighting based on entries to the 2010 US DOE next generation luminaire competition. Leukos 2012; 8: 237–259.

Houser

Wei

David

Krames

Shen

. Review of measures for light-source color rendition and considerations for a two-measure system for characterizing color rendition. Optics Express 2013; 21: 10393–10411.

Guo

Houser

. A review of colour rendering indices and their application to commercial light sources. Lighting Research and Technology 2004; 36: 183–199.

Wei

Houser

Allen

Beers

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

Wei

Houser

David

Krames

. Perceptual responses to LED illumination with colour rendering indices of 85 and 97. Lighting Research and Technology 2015; 47: 810–827.

10.

Hashimoto

Yano

Shimizu

Nayatani

. New method for specifying color-rendering properties of light sources based on feeling of contrast. Color Research and Application 2007; 32: 361–371.

11.

Judd

. A flattery index for artificial illuminants. Illuminating Engineering 1967; 62: 593–598.

12.

Davis

Ohno

. Color quality scale. Optical Engineering 2010; 49: 033602.

13.

Smet

KAG

Ryckaert

Pointer

Deconinck

Hanselaer

. Colour appearance rating of familiar real objects. Color Research and Application 2011; 36: 192–200.

14.

Thornton

. A validation of the color-preference index. Journal of the Illuminating Engineering Society 1974; 4: 48–52.

15.

Lin

Tsukitani

Noguchi

. Colour quality evaluation of natural objects based on the Feeling of Contrast Index. Lighting Research and Technology 2016; 48: 323–339.

16.

Jost-Boissard

Avouac

Fontoynont

. Assessing the colour quality of LED sources: naturalness, attractiveness, colourfulness and colour difference. Lighting Research and Technology 2015; 47: 769–794.

17.

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

18.

David

Fini

Houser

Ohno Y, Royer MP, Smet KAG, Wei M, Whitehead L . Development of the IES method for evaluating the color rendition of light sources. Optics Express 2015; 23: 15888–15906.

19.

Rea

Freyssinier

. Color rendering: a tale of two metrics. Color Research and Application 2008; 33: 192–202.

20.

Wei

Houser

. What is the cause of apparent preference for sources with chromaticity below the blackbody locus? Leukos 2016; 12: 95–99.

21.

Freyssinier JP, Rea MS. Class A color classification for light sources used in general illumination: Proceedings of the 13th International Symposium on the Science and Technology of Lighting Light Sources Troy, NY, June 24–29: 2012: pp. 337–338.

22.

Royer

. IES TM-30-15 is approved – now what? Leukos 2016; 12: 3–5.

23.

Wei M, Houser KW. Color discrimination of seniors with and without cataract surgery under illumination from two fluorescent lamp types: Proceedings of CIE 2012 Lighting Quality and Energy Efficiency, Hangzhou, China, 19–21 September 2012: 359–368.

24.

Houser

Wei

David

Krames

. Whiteness perception under LED illumination. Leukos 2014; 10: 165–180.

25.

David

. Color fidelity of light sources evaluted over large sets of reflectance samples. Leukos 2014; 10: 59–75.

26.

Wei M, Houser K, David A, Krames M. Effect of gamut shape on colour preference: Proceedings of CIE 2016 Lighting Quality and Energy Efficiency, Melbourne, Australia, March 3–5: 2016: pp. 32–41.

27.

Fotios

Houser

. Research methods to avoid bias in categorical ratings of brightness. Leukos 2009; 5: 167–181.

28.

Poulton

. Bias in Quantifying Judgements, Hove and London: Lawrence Erlbaum Associates, 1989.

29.

Dunn-Rankin

Knezek

Wallace

Zhang

. Scaling Methods, 2nd. Hove and London: Lawrence Erlbaum Associates, 2004.

30.

Quellman

Boyce

. The light source color preferences of people of different skin tones. Journal of the Illuminating Engineering Society 2002; 31: 109–116.

31.

Houser

Fotios

Royer

. A test of the S/P ratio as a correlate for brightness perception using rapid-sequential and side-by-side experimental protocols. Leukos 2009; 6: 119–138.

32.

Royer

Houser

. Spatial brightness perception of trichromatic stimuli. Leukos 2012; 9: 89–108.

33.

Wei M, Houser KW. Color preference under light stimuli characterized by a two-measure system: a pilot study. 2016 IES Research Symposium III – Light + Color, April 3–5 2016, Gaithersburg, MA, USA. (Poster).

34.

Luo

Cui

. Uniform colour spaces based on CIECAM02 colour appearance model. Color Research and Application 2006; 31: 320–330.

35.

Ohno Y. CQS v9.2 (32 samples, skin tones), Distributed within the IES Color Metrics Task Force, 2013.

36.

Lin

Wei

Smet

Tsukitani

Bodrogi

Khanh

. Colour preference varies with lighting application. Lighting Research and Technology 2015; Online first, DOI: 10.1177/1477153515611458. .

37.

Rea MS, Freyssinier JP. The Class A color designation for light sources: Proceedings of Experiencing Light 2012: International Conference on the Effects of Light on Wellbeing, Eindhoven, Netherlands, 12–13 November 2012.

38.

Smet

AKG

David

Whitehead

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

39.

Ashdown

Aviles

Bennett

, et al. Correspondence: In support of the IES method of evaluating light source colour rendition. Lighting Research and Technology 2015; 47: 1029–1034.