A source of white light with continuously tuned color rendition properties, such as color fidelity, as well as color saturating and color dulling ability has been developed. The source, which is composed of red (R), amber (A), green (G), and blue (B) light-emitting diodes, has a spectral power distribution varied as a weighted sum of “white” RGB and AGB blends. At the RGB and AGB end-points, the source has a highest color saturating and color dulling ability, respectively, as follows from the statistical analysis of the color-shift vectors for 1269 Munsell samples. The variation of the weight parameter allows for continuously traversing all possible metameric RAGB blends, including that with the highest color fidelity. The source was used in a psychophysical experiment on the estimation of the color appearance of familiar objects, such as vegetables, fruits, and soft-drink cans of common brands, at correlated color temperatures of 3000 K, 4500 K, and 6500 K. By continuously tuning the weight parameter, each of 100 subjects selected RAGB blends that, to their opinion, matched lighting characterized as “most saturating,” “most dulling,” “most natural,” and “preferential”. The end-point RGB and AGB blends have been almost unambiguously attributed to “most saturating” and “most dulling” lighting, respectively. RAGB blends that render a highest number of colors with high fidelity have, on average, been attributed to “most natural” lighting. The “preferential” color quality of lighting has, on average, been matched to RAGB blends that provide color rendition with fidelity somewhat reduced in favor of a higher saturation. Our results infer that tunable “color rendition engines” can validate color rendition metrics and provide lighting meeting specific needs and preferences to color quality.
©2012 Optical Society of America
So far, the assessment and control of the color rendition quality of illumination remains a problem with no ultimate solution. Ever since the introduction of advanced fluorescent lamps, the only widely recognized metric for ranking the color quality of artificial light sources has been the Color Rendering Index (CRI) recommended by the International Commission of Illumination (Commission Internationale de l’Éclairage, CIE) . The general CRI (Ra) is based on estimating color differences within the U*V*W* color space for 8 test color samples when a reference standard illuminant (blackbody or daylight phase) is replaced by the spectral power distribution (SPD) of a source under assessment (for sources with chromaticity somewhat different from the reference illuminant, a chromatic adaptation transform based on the von Kries hypothesis  is applied). This figure of merit increases toward 100 as the magnitude of the color differences is reduced to nearly zero, i.e. CRI measures color fidelity in respect of the reference illuminant. One of the drawbacks of the CRI metric is an inaccurate estimation of the color differences due to the lack of uniformity of the color space and the limited accuracy of the chromatic adaptation transform used . Numerous color spaces with improved perceptual uniformity have been suggested for the refinement of the CRI metric (see  and references therein). The most accurate color-difference estimations are based on the CIECAM02 color appearance model [3,5].
However starting with the introduction of CRI in 1964, it has been criticized not only for the reduced accuracy, but also for the disregard of subjective preferences to color rendition. Such preferences are related to the ability of light sources to make colors of illuminated objects appear “vivid” and easy to discriminate. To that end, numerous approaches that consider this aspect of color quality of lighting, such as the flattery index , color-discrimination index , color-preference index , visual clarity approach [9,10], color-rendering capacity , impression of colorfulness , index on feeling of contrast , and Gamut-Area Index (GAI)  (which measures the area of the polygon embraced by the chromaticity points of the same eight test color samples used in the general CRI), have been introduced. Basically, these approaches rely on the perceptually positive impact of the increased chroma of rendered colors (color saturating effect).
Solid-state lighting technology based on both narrow-band (direct-emission) and wide-band (phosphor converted) light-emitting diodes (LEDs) dramatically improved the versatility of composing the SPDs of light sources . This inspired numerous experiments on subjective preferences to color quality of illumination using various solid-state sources of light [12,16–23]. These experiments have shown that the CRI-based ranking of light sources contradicts the subjective visual ranking, when solid-state sources of light are compared to other light sources and to each other.
The failure of CRI stimulated a search for advanced color rendition metrics that would be able to quantify both color fidelity and subjective preferences to the color saturating effect. A reasonable approach is the two-metric system, which comprises the general CRI and GAI . An alternative approach is an attempt to quantify the ability of a light source to render object colors with high fidelity and to discriminate colors within an integral figure of merit, the general Color Quality Scale (CQS) [24,25]. This approach is based on the disregard of the color saturating component in the color shifts of 15 test color samples. It also contains numerous refinements, such as the use of more perceptually uniform CIELAB color space, improved scaling and averaging procedures, correlated color temperature (CCT) factor, and an advanced chromatic adaptation transform. The CQS metric is supplemented by additional scales, such as color fidelity scale (Qf, which is the refinement of the general CRI), gamut area scale (Qg, which is the refinement of GAI), and color preferences scale (Qp). The latter scale differs from the general CQS in having additional rewards for increased chroma. Both the two-metric system and CQS suffer from a small number of test color samples (8 and 15, respectively), which cover a negligible portion of the color space and make the assessment results dependent on the chosen sample set. The use of a large number of test color samples requires advanced metrics with the output format different from averaging color shifts and measuring gamut area.
An example of an advanced color rendition metric is the color rendition icon, which graphically condenses information on both the magnitude and direction of the color shifts for a large number (≥215) of test color samples . However, it might be too tricky to understand for the end-user and is difficult to apply when optimizing and designing light sources with required color rendition properties.
Another example of a color rendition metric with a large number test color samples is the statistical approach, which is based on the grouping of 1269 Munsell test color samples to particular indices depending on the magnitude and direction of the color shifts [27,28]. The samples that have color shifts smaller that the triple just-perceivable chromaticity and luminance difference  are considered as having no perceptually noticeable color distortions and are scored to the Color Fidelity Index (CFI). Once a color shift vector exceeds the triple difference, the sample is scored to another index depending on the type of color distortion determined by the direction of the color shift. The corresponding statistical indices are the Color Saturation Index (CSI), Color Dulling index (CDI), Hue Distortion Index (HDI), and Luminance Distortion Index (LDI). Eventually, the color rendition properties of a light source are presented in terms of these explicit and single-format indices (which are defined as percentages of the test color samples scored to each group).
However, the psychophysical assessment of the proposed multi-dimensional color rendition metric is difficult because of immense versatility in composing the SPDs of solid-state sources of light. Subjective side-by-side comparison of a large number of light sources is complex technically and exhausting for the subjects. Another problem is the lack of methods for the optimization of light sources in terms of subjective preferences to color quality. Commonly, the psychophysical experiments on color rendition preferences that involve solid-state lamps [12,16–19,21,22] rely on the ranking of light sources within a particular limited set rather than on finding the globally most preferred SPDs. Meanwhile, manual search of preferred SPDs [20,23,29] lacks a systematic approach. An important attempt to resolve this problem is the computational optimization of LED clusters using the memory color preference metric [22,30].
In this work, we present a concept of a solid-state source of light with continuously tunable color rendition properties. The source was used for the examination of the psychophysical validity of the general CRI, two-metric system, CQS, and statistical approach. This source allowed for a psychophysical optimization of the SPD for meeting subjective preferences to color quality. Some preliminary data dealing with this concept have been published in our previous papers [31–33].
2. Color rendition engineering
Our concept of color rendition engineering is based on the sensitivity of chromatic distortions of illuminated objects to the ratio of spectral power between the yellow and red regions of the spectra. The statistical analysis of color-shift vectors for a high number of test color samples [28,34] has shown that light sources that have balanced spectral components in the wavelength ranges of both 530-610 nm and beyond 610 nm can render most colors with small distortions of chroma and, therefore, provide high-fidelity illumination. Meanwhile lamps that lack spectral power in the 530-610 nm wavelength range render many colors with increased chroma, i.e. they exhibit the color-saturating effect. Alternatively, the lack of spectral power at wavelengths beyond 610 nm results in a decrease of chroma of many rendered colors, i.e. lamps with such SPDs are color-dulling. These regularities can also be traced in numerous estimations of the gamut area provided by different SPDs [24,25,35,36] and experiments on subjective assessment of different light sources [12,13].
Within such a concept, the simplest approach to a color rendition engine is a tetrachromatic solid-state lamp, composed of red (R), amber (A), green (G) and blue (B) LEDs. For a particular set of four primary LEDs, the three color mixing equations  show that the metameric SPDs of the RAGB lamp can be controlled by a single degree of freedom, such as the ratio of spectral power of the red and amber LEDs. By varying this ratio, the RAGB lamp can generate white light within an infinite number of metameric SPDs, each having different color rendition properties.
An appropriate set of LEDs can be selected from a range of standard commercial devices, such as offered by the Philips Lumileds Lighting Luxeon® Rebel family. These are, for instance, direct-emission royal-blue (peak wavelength 452 nm) and green (523 nm) InGaN LEDs, InGaN-based phosphor converted amber LED (589 nm), and direct-emission red AlGaInP LED (637 nm). Figure 1 shows the CIE 1931 chromaticity diagram with the color points of the four LEDs and white sources with different CCTs displayed. The white chromaticities fall within the AGB and RGB triangles, whereas the other two possible triangles (RAB and RAG) contain no white chromaticity points. This means that metameric SPDs of white light can be generated as an RGB blend or AGB blend or a linear combination of those with the controlled red-to-amber ratio (degree of freedom). For instance, the SPDs of the AGB and RGB blends, SAGB(λ) and SRGB(λ), respectively, can be combined into any tetrachromatic SPD as a weighted sum
The trichromatic AGB and RGB SPDs used in Eq. (1) are easy to design by solving conventional color mixing equations . For the calculated partial radiant fluxes generated by the constituent LEDs within the trichromatic modes, AAGB, GAGB, and BAGB and RRGB, GRGB, and BRGB, respectively, the partial radiant fluxes of the tunable RAGB cluster, RRAGB, ARAGB, GRAGB, and BRAGB, are given by the following linear functions:
The SPD of the RAGB lamp can be dynamically tuned by varying the average driving current of individual LEDs within a cluster in accordance with Eqs. (2).
We developed a prototype tunable RAGB light source containing 7 identical clusters of 4 colored LEDs. A 10-bit pulse-width modulation electronic circuit controlled by a computer varied partial fluxes generated by each group of the LEDs. The software also allowed for varying CCT and maintaining constant net luminous flux while tailoring the SPD. To avoid chromatic adaptation problems, the SPDs were tailored to bring the chromaticity point as close as possible to the blackbody locus (for CCT below 5000 K) or to the CIE daylight locus (for CCT above 5000 K). The LEDs were mounted on a massive heat sink cooled by air. The spectra were measured and calibrated using a CCD spectrometer with a fiber-optic probe (Hamamatsu model PMA-11) under the conditions of the preheated heat sink. The source was mounted on top of an experimental cabinet (Fig. 2 ) over a color-mixing ceiling made of Plexiglas and provided an illuminance of about 700 lx at the bottom surface of the cabinet with unevenness better than 4%. During the experiments described below, the chromaticity point of the LED cluster measured at the bottom surface of the cabinet was maintained within 3 MacAdam ellipses (~0.006 xy distance from the required chromaticity point).
Figures 3(a), (b), and (c) depict the variation of the radiant fluxes generated by each of the four LED groups with the weight parameter σ for CCTs of 3000 K, 4500 K, and 6500 K, respectively. In accordance with Eqs. (2), the flux of the red component linearly decreases from the peak value to zero and the flux of the amber component increases from zero to the peak value as the weight increases from 0 to 1. The flux of the green component linearly decreases with weight between the end-point values, whereas the flux of the blue component stays at an almost constant value. With increasing CCT, the fluxes of the red and amber components decrease, whereas those of the green and blue components increase.
Figures 3(d), (e), and (f) show the variation of the general CRI, GAI, the general CQS, and CQS color preference scale Qp (here designated as CPS) with the weight parameter for the three values of CCT. With the variation of AGB vs. RGB weight from 0 to 1, the gamut area (GAI) gradually decreases, whereas the general CRI, general CQS, and CPS vary non- monotonously, i.e. they increase, pass through maximal values, and then decrease. These properties of our tunable RAGB lamp clearly indicate that neither the general CRI nor general CQS can be used as single figures of merit, since they can produce same values that correspond to different gamut areas (e.g., at 4500 K the general CRI has a value of 80 for two different SPDs with GAIs of about 78 and 101). The use of the two-metric (CRI/GAI) system  resolves this problem. Similarly, two distinct scales are to be used in the CQS metric (color fidelity scale and gamut area scale ) instead of the integration of the two color rendition characteristics within the general CQS.
Figures 3(g), (h), and (i) show the statistical color quality indices CFI, CSI, CDI, and HDI as functions of weight for the same three values of CCT. The CFI (percentage of color test samples rendered with high fidelity) has peak values of 87%, 79%, and 76% at weights of 0.69, 0.73, and 0.80 for CCT of 3000 K, 4500 K, and 6500 K, respectively. Interestingly, these peak values are very close to the peaks of the general CRI, which also measures the fidelity of color rendition. However, the variation of CFI is much more pronounced due to the distinct discrimination of the color-shift vectors. (In the statistical metric, color-shift vectors that show perceivable color distortions do not score to the fidelity index.) The highest values of CSI and CDI (percentages of test color samples rendered with increased and decreased chroma, respectively) are attained at the end-points of σ = 0 and σ = 1, respectively. At σ = 0, CSI equals 80%, 72%, and 67% for CCT of 3000 K, 4500 K, and 6500 K, respectively. At σ = 1, CDI equals 64%, 51%, and 39%, respectively. As the weight parameter deviates from the corresponding end-point values, both indices rapidly decrease and drop to marginal values of 1 to 5%, when the highest values of CFI are attained. At the opposite end-points, the CSI and CDI nearly completely vanish (dropping below 2%). Again, these two statistical indices show much more pronounced behavior than their gamut-area counterpart (GAI) due to the threshold-like discrimination of the color-shift vectors, which avoids scoring unless the vector contributes to a perceptually perceived alteration of chroma.
3. Psychophysical experiment
A psychophysical experiment was carried out to demonstrate the use of the tunable RAGB source for the validation of color rendition metrics and for the optimization of the tetrachromatic SPDs to meet subjective preferences to color quality of illumination. In our experiment, the subjects continuously tuned the color rendition properties of the source to select a spectrum that, to their opinion, met particular characteristics. This is different from previous studies [12,16–19,21,22], where the subjects ranked light sources within an unsystematic set.
During the experiment, subjects viewed a set of familiar objects [8,19,22], such as vegetables and fruits (lettuce, tomato, banana, lemon, light-green grapes, nectarine, and orange) and soft-drink aluminum cans of common brands: Coca-Cola® (red), Sprite® (yellow/green/blue), and Pepsi-Cola® (blue/red) (see Fig. 2). The objects were placed within a cabinet made of neutral gray matted plastic with dimensions of 46 cm by 46 cm by 46 cm and viewed from a distance of 60 to 70 cm. The cabinet was installed in a dark room. A laptop computer was used for the continuous control of the weight parameter of the SPD by incrementing it in steps of 0.05. Repetitive clicking the mouse pointer on either of two icons displayed on the monitor resulted in the cyclical incrementing of weight (from 0 toward 1, then back from 1 toward 0, and so on). Clicking the pointer on the other icon resulted in a reverse variation of the weight parameter. This allowed for precisely controlling the source by forward-backward tuning, without latching onto any of the end-points.
100 subjects (not acquainted with the details of color rendition problem) took part in the experiments. Of those 41 were females and 59 were males; the age ranged from 14 to 78 (with the average age of 27). All subjects were Caucasians with East-European cultural background. Prior to the experiment they were screened against color vision deficiencies using Stilling’s pseudoisochromatic plate tests. Each subject was verbally instructed for about 3 minutes and had about 5 minutes to adapt to the scene and to perform a trial tuning of the source. Afterward, the subjects were asked to select four options of lighting, which were verbally characterized as “most saturated,” “most dull,” “most natural,” and “preferential.” Each subject performed the selection of the four options in the above sequence from 21 circularly varied ones at three CCTs of 3000 K, 4500 K, and 6500 K (for each subject, the CCT was changed in a random sequence). The time allocated for a selection was not limited; however selections of “most saturated” and “most dull” lighting were accomplished, on average, in about half a minute. The “most natural” and “preferential” selections took somewhat longer, typically about one minute, on average. As soon as the option was selected, the experimenter recorded the weight parameter, which the subject did not see. Since the experiment lasted for several weeks, fruits and vegetables were being replaced by those having similar appearance each 2-3 days.
The points in Figs. 3(j), (k), and (l) display the percentage of 21 weights corresponding to the four subjectively selected options of color quality. The lines show the distributions of the data approximated by Gaussian functions. The open circles with the horizontal bars show the mean values with their 95% confidence intervals for the weight for “most natural” and “preferential” selections. The obtained distributions of selection rate exhibit the following trends:
- • the subjective identification of lighting with the highest distortions of chroma (“most saturated” and “most dull”) are highly correlated with the highest CSI and CDI at the RGB and AGB end-points, respectively (the end-points were exactly identified in 61-75% trials);
- • the identification of “most natural” and “preferential” lighting resulted in considerably wider distributions;
- • the average AGB vs. RGB weights subjectively attributed to “natural” lighting (0.74 ± 0.07, 0.78 ± 0.05, and 0.73 ± 0.09 at CCTs of 3000 K, 4500 K, and 6500 K, respectively) agree with those estimated for the highest CFI (0.69, 0.73, 0.80, respectively) within the 95% confidence intervals;
- • the average AGB vs. RGB weights subjectively attributed to “preferential” lighting (0.53 ± 0.08, 0.55 ± 0.06, and 0.51 ± 0.10 at CCTs of 3000 K, 4500 K, and 6500 K, respectively) are noticeably shifted (by about 0.22) toward higher saturation in respect of those attributed to “most natural” lighting (in over 70% of the trials, the preference was shifted to higher saturation, whereas a shift to lower saturation was made only in about 20% of the trials).
The above experiment on continuous traversing metameric SPDs of the RAGB source provides extended data on the psychophysical validation of color rendition metrics. In order to complete such a validation, we should consider the spectral and color-shift properties of the tetrachromatic blends corresponding to the peak selection rates for the illumination conditions with different characteristics.
Figure 4 shows the SPDs of the four RAGB blends for a CCT of 4500 K corresponding to the highest subjective selection rate for the color appearance characterized as “most saturated” (a), “preferred” (c), “most natural” (e), and “most dull” (g), respectively. Figures 4(b), (d), (f), and (h) show the corresponding distributions of the color-shift vectors in respect of a blackbody radiator estimated for 218 Munsell samples of value /6, respectively. These distributions are presented within the a*−b* chromaticity plane of the CIELAB color space; arrows schematically show the vectors that are estimated to have perceptual noticeable chromaticity distortions ; circles correspond to the color samples rendered with high fidelity. For 3000 K and 6500 K, the SPDs and color-shift vector distributions have similar properties.
The experimental data (Figs. 3(j), (k) and (l)) show that the subjective tuning of the RAGB source for matching the most noticeable color saturating and color dulling conditions was the easiest and least ambiguous task to accomplish. Out of all metameric SPDs, the end-point RGB (σ = 0, Fig. 4(a)) and AGB (σ = 1, Fig. 4(c)) blends were almost exactly identified with the “most saturated” and “most dull” appearance of familiar objects. These two blends render the highest statistical percentage of a large number of colors with increased and decreased chroma (the largest CFI and CDI), respectively. The color-saturating RGB blend, which features a low spectral power in the 540-610 nm range (Fig. 4(a)), results in numerous color-shift vectors with a perceptually noticeable component directed toward increased chroma (Fig. 4(b)). Alternatively, the color-dulling AGB blend (Fig. 4(g)), which features low spectral power beyond 610 nm, results in numerous vectors exhibiting perceptually noticeable reduction of chroma (Fig. 4(h)). With a typical consistency of subjective color quality assessment of about 70% , the obtained rate of recognizing the end-point color distortion patterns (while continuous tuning the SPDs) indicates that the subjective feeling of color saturation is highly correlated with the number of colors that have distorted chroma. The end-points are well recognized despite a large number of colors with perceptually noticeable distortions of hue (typically, over 60% for color-saturating RGB blend and about 40% for color-dulling AGB blend).
The measure of gamut area for a small number of test color samples (GAI) is also in line with the selection rates for “most saturated” and “most dull” lighting. For a particular set of continuously varied blends, such as generated by the RAGB color rendition engine, GAI reasonably follows the variation of the color-shift vector distributions, despite the limited accuracy of the U*V*W* color space used. However, GAI lacks global correlation with the statistical metric, which is more comprehensive. The ranking of light sources using GAI might be different from that based on CSI and CDI .
The data presented in Figs. 3(j), (k) and (l) shows a remarkable agreement of the subjective selection rates for the “most natural” appearance of familiar objects with the highest values of statistical CFI (p < 0.05). Such an agreement is clearly revealed, despite a rather wide distribution of the selections. The wide distribution is probably due to fact that when judging on the “most natural” appearance of the objects, the subjects relied on their memory and on the individual way of judging rather than on direct visual impression as in the cases of “most saturated” and “most dull” specifications. Figure 4(e) shows the tetrachromatic SPD corresponding to the highest selection rate for “most natural” illumination (σ = 0.78). The blend has a balanced spectral power within the spectral ranges of 540-610 nm and beyond 610 nm. The distribution of the color-shift vectors (Fig. 4(f)) shows that most test color samples have no perceptually noticeable distortions of chroma (open circles) and those that do have very small distortions compared to the RGB and AGB end-points. These data prove that the subjective feeling of color rendition “naturalness” can be associated with a low number of chromatic distortions perceived while comparing the observed appearance of objects with memorized colors.
The subjective perception of color “naturalness” represented by the distributions shown in Figs. 3(j), (k), and (l) agrees with the variation of the general CRI (Figs. 3(d), (e), and (f), respectively), which also measures the color fidelity of illumination. Such an agreement of the selection rate vs. weight dependence with the peak of the general CRI occurs despite the limited accuracy of the U*V*W* color space. The reason is that when color shifts shrink, any figure of merit measuring color fidelity in terms of reduced shifts arrives at its highest value, irrespectively of the color space used (the general CRI attains values in the range of 93 to 95 in our case). However, the general CRI may miss many distorted colors, in particular of red hues, because of a small number of test color samples used. The enhanced color fidelity determined by using a large number of test color samples requires the SPDs with red components shifted to longer wavelengths  in respect of those in the SPDs obtained through maximizing the general CRI .
One of the most interesting results of our psychophysical research is the possibility of establishing the subjectively preferred RAGB blends (note that this applies to a particular scene and to a particular group of subjects). Figure 4(c) shows the SPD of the blend corresponding to the highest selection rate for “preferred” appearance of familiar objects (σ = 0.55). It differs from that selected for “most natural” lighting (Fig. 4(e)) by a larger red-to-amber spectral power ratio. The distribution of the color-shift vectors for the “preferred” blend indicates a reduced number of the test color samples rendered with high fidelity and an increased number of the test color samples having increased chroma. However, the characteristics of “preferred” illumination are very far from those of highly saturating RGB end-point. This means that neither statistical CSI nor gamut-area or chroma-increase based indices, such as GAI and CQS gamut-area index, can serve as the indicators of the color quality preference. The reason is that, in contrast to fluorescent lamps used in early experiments , the color saturating ability of an RGB LED cluster can be so high that the colors of illuminated objects become oversaturated and lose their subjective attractiveness. Another reason for the observed shift of the “preferred” selections from highly saturating RGB blend to that providing higher fidelity might be the aforementioned large number of colors with perceptually noticeable hue distortions [27,28]. The CQS color-preference scale has a maximum almost coinciding with the highest selection rate for “preferred” illumination. However, Fig. 4(e) shows that CQS Qp is very insensitive to the increase of the color-saturating ability of the RAGB source and does not distinguish distinctively between sources that render colors with high fidelity, increased saturation, and subjective preference.
Therefore, the results of our study indicate that a sole figure of merit quantifying preferences to color rendition quality of illumination is unlikely to be feasible. A partial solution is to quantify the preferred color rendition by a ratio of two indices that measure color saturating ability and color fidelity, respectively. Depending on the metric, such ratios as GAI/Ra, Qg/Qf, or CSI/CFI, might be useful measures of the color-preference effect. For instance, our experimental data (for a particular scene and a particular group of subjects) show that the CSI/CFI ratio characteristic of the highest rate of “preferred” selections is about 1.1, 1.2, and 1.6 for CCTs of 3000 K, 4500 K, and 6500 K, respectively. However, individual preferences to color quality might depend on many factors, such as the character of the SPD of the source, the luminance and chromatic environment of specific scenes (merchandise, architectural, entertainment, museum, temple, medical, landscape, etc.), as well as age, gender, racial and cultural background, and the observer mood.
Therefore, we propose using a color rendition engine (such as demonstrated above) as an ultimate solution for meeting individual preferences to color quality. Such color rendition engines can be based on several types of LED clusters. For instance, at the color-dulling end-point, a dichromatic (yellow-blue) phosphor converted LED or an appropriate YB LED cluster, which have pronounced color-dulling properties [27,28], can be used instead of the AGB LED cluster . In our estimate, such WRGB and RYGB lamps have the properties of color rendition tuning similar to those of the RAGB cluster. However, they lack the ability to control CCT.
We have developed a versatile solid-state source of white light with tunable color rendition properties and used it as a color rendition engine. The SPD of the source is a weighted sum of RGB and AGB white blends designed for a particular CCT. Varying the AGB vs. RGB weight parameter,σ, makes the RAGB lamp to continuously traverse all possible metameric tetrachromatic blends.
At the RGB end-point (σ = 0), the source renders the largest number of colors with perceptually noticeable increased chroma out of 1269 test color samples and provides the largest gamut area for a small number test color samples, such as used in the CRI metric. At the AGB end-point (σ = 1), it attains the largest number of colors with perceptually noticeable decreased chroma and the smallest gamut area, respectively. With increasing the weight parameter from 0 to 1, the number of colors rendered with increased chroma and the gamut area gradually decrease, while the number of colors rendered with decreased chroma increase. At particular values of the weight parameter, the source arrives at the blends that provide the highest color fidelity measured as the largest number of 1269 colors rendered without perceptually noticeable color distortions. The blends with the highest values of the general CRI, general CQS, and CQS color-preference scale can be generated as well.
A psychophysical investigation on the estimation of the color appearance of familiar objects, such as vegetables, fruits, and soft-drink cans of common brands, has been carried out for CCTs of 3000 K, 4500 K, and 6500 K. By continuously forward-backward tuning the weight parameter, each of 100 subjects selected RAGB blends matching, in their opinion, lighting characterized as “most saturating,” “most dulling,” “most natural,” and “preferential”. The analysis of the selection rates indicated that the subjects easily recognized color saturating (large gamut area) and color dulling (low gamut area) end-point blends, probably due to the direct visual impression. The selection rate for “most natural” illumination spanned over a wider distribution of RAGB blends, probably due to the comparison of the observed colors with those contained in memory and due to the individual way of judging. However, the average AGB vs. RGB weights subjectively attributed to “most natural” lighting agreed with those estimated as providing the highest number of colors rendered with high fidelity. The average weights subjectively attributed to “preferential” lighting are somewhat shifted toward a higher saturation in respect of those attributed to “most natural” lighting.
Both the variation of various color rendition indices while tuning our color rendition engine and the results of the psychophysical experiment suggest that no single figure of merit is capable of distinctively quantifying the color quality of illumination. At least two figures of merit, one for the estimation of color fidelity and the other for the estimation of color saturating effect, are required to reasonably rate the color rendition properties of light sources, including color preference. We suggest that the most comprehensive and highly distinctive assessment of color quality of illumination is provided by the statistical approach, which sorts a large number of test color samples depending on the presence and type of perceptually noticeable color distortions. This approach rates light sources in terms of single-format indices for color fidelity, color saturating and color dulling ability. However, the context-dependent and individual preferences to color quality of illumination are difficult to unambiguously scale in numerical indices; and these preferences are to be immediately met through the use of light sources with tunable color rendition properties, such as the color rendition engine demonstrated above.
The work at VU was funded from the Research Council of Lithuania by a grant (No. MIP-73/2010). The work at RPI was supported primarily by the Engineering Research Centers Program (ERC) of the National Science Foundation under NSF Cooperative Agreement No. EEC-0812056 and in part by New York State under NYSTAR contract C090145.
References and links
1. Commission Internationale de l’Eclairage, “Method of measuring and specifying colour rendering properties of light sources,” Pub. CIE 13.3, (1995).
2. G. Wyszecki and W. S. Stiles, Color Science. Concepts and Methods, Quantitative Data and Formulae (Wiley, 2000).
3. N. Sándor and J. Schanda, “Visual color rendering based on color difference evaluations,” Lighting Res. Tech. 38(3), 225–239 (2006). [CrossRef]
4. M. R. Luo, G. Cui, and C. Li, “Uniform color spaces based on CIECAM02 colour appearance model,” Color Res. Appl. 31(4), 320–330 (2006). [CrossRef]
5. C. Li, M. R. Luo, G. Cui, and C. Li, “Evaluation of the CIE colour rendering index,” Color. Technol. 127(2), 129–135 (2011). [CrossRef]
6. D. B. Judd, “A flattery index for artificial illuminants,” Illum. Eng. 62, 593–598 (1967).
8. W. A. Thornton, “A validation of the color-preference index,” J. Illum. Eng. Soc. 4, 48–52 (1974).
9. S. M. Aston and H. E. Belichambers, “Illumination, color rendering, and visual clarity,” Lighting Res. Tech. 1(4), 259–261 (1969). [CrossRef]
10. K. Hashimoto and Y. Nayatani, “Visual clarity and feeling of contrast,” Color Res. Appl. 19(3), 171–185 (1994). [CrossRef]
12. Y. Nakano, H. Tahara, H. Suehara, J. Kohda, and T. Yano, “Application of multispectral camera to color rendering simulator,” in Proceedings of the 10th Congress of the International Colour Association AIC Colour 05, J. L. Nieves and J. Hernández-Andrés, eds. (AIC, 2005), pp. 1625–1628.
13. K. Hashimoto, T. Yano, M. Shimizu, and Y. Nayatani, “New method for specifying color-rendering properties of light sources based of feeling of contrast,” Color Res. Appl. 32(5), 361–371 (2007). [CrossRef]
14. M. S. Rea and J. P. Freyssinier-Nova, “Color rendering: A tale of two metrics,” Color Res. Appl. 33(3), 192–202 (2008). [CrossRef]
15. M. Shur and A. Žukauskas, “Solid-state lighting: Toward superior illumination,” Proc. IEEE 93(10), 1691–1703 (2005). [CrossRef]
16. N. Narendran and L. Deng, “Color rendering properties of LED sources,” Proc. SPIE 4776, 61–67 (2002). [CrossRef]
17. I. Shakir and N. Narendran, “Evaluating white LEDs for outdoor landscape lighting application,” Proc. SPIE 4776, 162–170 (2002). [CrossRef]
18. Commission Internationale de l’Eclairage, “Colour rendering of white LED sources,” Pub. CIE 177, 2007.
19. S. Jost-Boissard, M. Fontoynont, and J. Blanc-Gonnet, “Perceived lighting quality of LED sources for the presentation of fruit and vegetables,” J. Mod. Opt. 56(13), 1420–1432 (2009). [CrossRef]
20. S. Boissard and M. Fontoynont, “Optimization of LED-based light blendings for object presentation,” Color Res. Appl. 34(4), 310–320 (2009). [CrossRef]
21. M. S. Rea and J. P. Freyssinier, “Color rendering: Beyond pride and prejudice,” Color Res. Appl. 35(6), 401–409 (2010). [CrossRef]
22. K. A. G. Smet, W. R. Ryckaert, M. R. Pointer, G. Deconinck, and P. Hanselaer, “Memory colours and colour quality evaluation of conventional and solid-state lamps,” Opt. Express 18(25), 26229–26244 (2010). [CrossRef] [PubMed]
23. K. Smet, W. R. Ryckaert, M. R. Pointer, G. Deconinck, and P. Hanselaer, “Colour appearance rating of familiar real objects,” Color Res. Appl. 36(3), 192–200 (2011). [CrossRef]
24. W. Davis and Y. Ohno, “Toward and improved color rendering metrics,” Proc. SPIE 5941, 594111 (2005).
25. W. Davis and Y. Ohno, “Color quality scale,” Opt. Eng. 49(3), 033602 (2010). [CrossRef]
26. P. van der Burgt and J. van Kemenade, “About color rendition of light sources: The balance between simplicity and accuracy,” Color Res. Appl. 35, 85–93 (2010).
27. A. Žukauskas, R. Vaicekauskas, F. Ivanauskas, H. Vaitkevičius, P. Vitta, and M. S. Shur, “Statistical approach to color quality of solid-state lamps,” IEEE J. Sel. Top. Quantum Electron. 15(6), 1753–1762 (2009). [CrossRef]
28. A. Žukauskas, R. Vaicekauskas, and M. S. Shur, “Color rendition properties of solid-state lamps,” J. Phys. D Appl. Phys. 43(35), 354006 (2010). [CrossRef]
29. W. Davis, J. L. Gardner, and Y. Ohno, “NIST facility for color rendering simulation,” in Proceedings of the 10th Congress of the International Colour Association AIC Colour 05, J. L. ONieves and J. Hernández-Andrés, eds. (AIC, 2005), pp. 519–522.
31. A. Žukauskas, R. Vaicekauskas, A. Tuzikas, and M. S. Shur, “Solid-state lighting with tailored colour quality,” in Proceedings of the 12th International Symposium on Science and Technology of Light Sources and 3rd International Conference on White LEDs and Solid State Lighting LS-WLED 2010, M. Haverlag, G. M. W. Kroesen, and T. Taguchi, eds. (FAST-LS, 2010), pp. 363–364.
32. A. Žukauskas and R. Vaicekauskas, “LEDs in lighting with tailored color quality,” Int. J. High Speed Electron. Syst. 20(02), 287–301 (2011). [CrossRef]
33. A. Žukauskas, R. Vaicekauskas, P. Vitta, A. Tuzikas, and M. Shur, “Statistical approach to color rendition properties of solid-state light sources,” Proc. SPIE 8123, 81230X, 81230X-9 (2011). [CrossRef]
35. Y. Ohno, “Spectral design considerations for white LED color rendering,” Opt. Eng. 44(11), 111302 (2005). [CrossRef]
36. V. Viliūnas, H. Vaitkevičius, R. Stanikūnas, A. Švegžda, and Z. Bliznikas, “LED-based metameric light sources: Rendering the colours of objects and other colour quality criteria,” Lighting Res. Tech. 43(3), 321–330 (2011). [CrossRef]
37. A. Žukauskas, R. Vaicekauskas, F. Ivanauskas, H. Vaitkevičius, and M. S. Shur, “Rendering a color palette by light-emitting diodes,” Appl. Phys. Lett. 93(2), 021109 (2008). [CrossRef]