Color Rendering of Light Sources
Color Rendering vs. Color QualityThe quality of object color under artificial lighting is an important aspect of the value of the light sources, particularly to consumers. Assessing and quantifying this dimension of lamps is complicated.
The attribute of color rendering of light sources is often interpreted as indicating object color quality. However, color rendering is actually defined as the “effect of an illuminant on the color appearance of objects by conscious or subconscious comparison with their color appearance under a reference illuminant” [Ref. 1]. Color rendering only refers to color fidelity, the accurate representation of object colors compared to those same objects under a reference source, and does not include other aspects of color quality, such as chromatic discrimination and color preference. The CIE color rendering index (CRI) [Ref. 2] is the only internationally-accepted metric for assessing the color rendering performance of light sources.
Color Rendering Index (CRI)In the calculation of the CRI, the color appearance of 14 reflective samples is simulated when illuminated by a reference source and the test source. The reference source is a Planckian radiator (if below 5000 K) or a CIE Daylight source (if at or above 5000 K), matched to the correlated color temperature (CCT) of the test source. After accounting for chromatic adaptation with a Von Kries correction, the difference in color appearance ΔEi for each sample between the test and reference light sources is computed in CIE 1964 W*U*V* uniform color space. The special color rendering index (Ri ) is calculated for each reflective sample by:
The general color rendering index (Ra ) is simply the average of Ri for the first eight samples, all of which have low to moderate chromatic saturation:
A perfect score of 100 represents no color differences in any of the eight samples under the test and reference sources.
Figure 2. The eight color samples used in the calculation of Ra.
Shortcomings of the CRIThe CRI has a number of problems, particularly when applied to LEDs or when used as an indicator of color quality. The uniform color space used to calculate color differences is outdated and no longer recommended for use. The red region of this color space is particularly non-uniform. Instead, the CIE currently recommends CIE 1976 L*a*b* (CIELAB) and CIE 1976 L*u*v* (CIELUV) [Ref. 3] for calculating color differences. Additionally, the chromatic adaptation transform is considered inadequate. The Von Kries chromatic adaptation correction used in the CRI has been shown to perform poorer than other available models, such as the CMCCAT2000 (the Colour Measurement Committee’s chromatic adaptation transform) and the CIE CAT02 (the CIE’s chromatic adaptation transform) [Ref. 4].
The CRI method specifies that the CCT of the reference source be matched to that of the test source, which assumes complete chromatic adaptation to the chromaticity of the light source. This assumption fails at extreme CCTs, however. For example, a 2000 K (very reddish) blackbody source achieves Ra = 100, as does a daylight spectrum of 20,000 K (very bluish). However, neither of these sources renders colors well.
None of the eight reflective samples used in the computation of Ra are highly saturated. This is problematic, especially for the peaked spectra of white LEDs. Color rendering of saturated colors can be very poor even when the Ra value is good. Further, by optimization of lamps’ spectra to the CRI, Ra values can be made very high while actual color rendering is much poorer. This problem exists because too few samples are used in the calculation of Ra, and they are of too low chromatic saturation.
The eight special color rendering indices are simply averaged to obtain the general color rendering index. This makes it possible for a lamp to score quite well, even when it renders one or two colors very poorly. LEDs are at an increased risk of being affected by this problem, as their peaked spectra are more vulnerable to poor rendering in only certain areas of color space.
Finally, the very definition of color rendering is flawed for use when one is interested in the overall color quality of a light source. Color rendering is a measure of only the fidelity of object colors under the illuminant of interest and any deviations of object color appearance from under a blackbody source is considered bad. Due to this constraint, all shifts in perceived object hue and saturation result in equal decrements in CRI score. In practical application, however, increases in chromatic saturation, observed when certain sources illuminate certain surfaces, is considered desirable. Increases in saturation yield better visual clarity and enhance perceived brightness [Ref. 5]. It is proposed that the absolute focus on color fidelity of the CRI is flawed and a more general metric of color quality be considered.
Further details on the analyses of CRI for solid-state sources are found in references [Ref. 8, Ref. 9, Ref. 10].
Color Quality Scale (CQS)To solve the problems of the CRI for solid-state light sources mentioned above, and to meet the new needs in the lighting industry and consumers for communicating color quality of all lighting products, a Color Quality Scale (CQS) is being developed at NIST, which evaluates several aspects of the quality of the color of objects illuminated by a light source. The extensive description of the CRI was provided because, rather than inventing an entirely new approach to the metric, much inspiration was taken from the CRI. Borrowing from aspects of the CRI that are successful, the CQS incorporates important modifications to overcome its shortcomings and focuses on a broader definition of color quality. The CQS is being developed with close contacts with the lighting industry and the CIE to be proposed as a future international standard.
The set of reflective samples tested is different from those used in the calculations of the CRI. Fifteen saturated Munsell samples are used in the CQS, with the following hue value/chroma: 7.5 P 4 / 10, 10 PB 4 / 10, 5 PB 4 / 12, 7.5 B 5 / 10, 10 BG 6 / 8, 2.5 BG 6 / 10, 2.5 G 6 / 12, 7.5 GY 7 / 10, 2.5 GY 8 / 10, 5 Y 8.5 / 12, 10 YR 7 / 12, 5 YR 7 / 12, 10 R 6 / 12, 5 R 4 / 14, and 7.5 RP 4 / 12. They were selected to have the highest chroma, span the entire hue circle in approximately even spacing, and be commercially available. Figure 1 shows these samples (bottom row) as well as the eight samples used in the calculation of Ra (top row) when illuminated by a daylight-like source (D65). This representation may be inaccurate due to the properties of the viewing display.
Figure 4: The 15 samples used by the Color Quality Scale (CQS).
The uniform object color space also differs from that used in the CRI. The 1964 W*U*V* object color space is obsolete, and is very nonuniform: color differences are extremely exaggerated in the red region and suppressed in yellow and blue regions. So, when calculating the CQS, CIE 1976 L*a*b* (“CIELAB”) is used, as it is currently recommended for use by the CIE and is considered to be reasonably uniform.
One of the major deviations that the CQS takes from the formal definition of color rendering is evident in the saturation factor. The CRI penalizes lamps for shifts in hue, chroma (chromatic saturation), and lightness, in any direction, of the reflective samples under the test source (compared to under the reference source). While a decrease in chroma always has negative effects, an increase in the chroma of objects is considered desirable in many cases. Increases in chroma yield better visual clarity and enhance perceived brightness [Ref. 5]. These are positive effects and are generally preferred, though they cause deviations in color fidelity (compared to reference). In the CQS, lamps are not penalized for increasing object chroma relative to the reference source, though their scores are also not increased. The net result is that a lamp’s score is only penalized for hue shifts, lightness shifts, and reductions in chroma. This is a way to take color preference (and possibly also color discrimination) into account in the CQS. For example, in Figure 5"A" when the chroma increases under the test illuminant (with no change in hue), there is no change in score, 5"B" when the chroma decreases under the test illuminant, the score is decreased, and 5"C" when the chroma increases and the hue shifts, the score is decreased for the hue shift but not decreased for the increase in chroma.
Further details on the CQS are found in reference [Ref. 11].
In the CRI, the CCT of the reference source is matched to that of the test source. Therefore the CRI score is perfect (100) for reference sources of any CCT. Actual color rendering, however, is degraded at extremely low or high CCTs. This is a problem with the way the reference source is defined in the current metric, and is one of the most difficult problems to address. The perfect solution to this problem would require thorough understanding of chromatic adaptation. Such investigations have not been conducted yet, but a temporary solution has been developed. Though the CCT of the reference source is also matched to that of the test in the CQS, a multiplication factor is introduced. This CCT factor is determined based on the gamut area in CIELAB space for the 15 samples under the reference source for each CCT, as shown in Table 1. It is assumed that the color rendering performance of the reference source degrades as the gamut area decreases. The multiplication factor, as listed in the table, is the ratio of gamut area of the particular CCT with the gamut area for 6500 K. With this normalization, the multiplication factors at certain CCT ranges (e.g., 4000 K) give values slightly higher than 1, but these are truncated to 1, so that the CQS score will never be higher than 100. The exact effect of CCT on color quality is difficult to quantify, but this method offers at least a temporary solution for sources of extremely low or high CCT.
Testing the CQSMany computational simulations have been performed and, at the level of subjective visual impression, appear to confirm the ideas used in the CQS. For details and examples, see Ref. 8.
A series of thorough and well-controlled vision experiments are necessary to test, improve upon, and validate the computational analyses, however. Experiments testing observers’ chromatic discrimination and absolute hue perception of illuminated objects will be complemented by subjective rankings of naturalistic scenes. Since the CQS is intended to be a metric of overall color quality, the data from several types of experiments will be used to assess and improve its performance. A new vision science laboratory is being built at NIST, with experiments testing CQS taking highest priority.
Return to Vision Science