Ra, CQS, GAI, CRI2012, FSI... Is it even possible to express the colour fidelity of artificial light with a single number?

Why is it necessary to evaluate the colour rendering of light sources? "It is essential for visual performance and a sense of overall and mental well-being that the colours of objects and human skin in a given environment are rendered naturally, faithfully and in such a way that people look attractive and healthy." [1].

The standard [1] for the purpose of visual performance (correct differentiation of differences and colour matching at work) specifies the minimum required values of the general colour rendering index Ra, in most cases 80. For example, for colour and quality control, colour grading and art work in crafts and industry, hairdressing, for many areas in healthcare, art rooms in art schools or dressing rooms in theatres, a Ra of at least 90 is required.

The administration of human skin tone is essential for medical diagnosis. Examples include cyanosis (turning blue with inadequate oxygenation) or jaundice. Both warning colours can be "masked" by the absence of long-wave red or yellow radiation in the light of three-band fluorescent lamps. Natural illumination of the face is also of great importance in the extra-verbal reception of non-verbal signals (body language) in interpersonal communication [2].

The associations between shape and colour are stored in a person's colour memory. The colour of healthy human skin (especially the face), road signs, warning signs, company logos, uniforms or even the design of mailboxes facilitate orientation and decision-making by linking the perception of shape and colour. When these 'memory' colours are distorted, one quickly gets the feeling that something is wrong.

The appearance of food is a special case. Our brains know what a ripe orange, a good carrot or a fresh ham should look like. The lighting in supermarkets makes these expected "memory" colours of fruit, vegetables, bread or meat stand out. Sometimes, however, it covers up their shortcomings. Similarly, textiles look better in light with a good rendering of rich colours, which is exploited by clothing, fabric and carpet retailers.

The feeling of general and mental well-being is supported in all respects by a well-designed lighting system that provides illuminance and chromaticity temperature in the pleasant area of the Kruithof diagram, and a colour rendering appropriate to the activity or relaxation being performed. Warmer light has a calming effect and creates a feeling of security, while cooler tones emphasise the feeling of lightness and cleanliness of the space and encourage activity. A sense of colour balance and a sufficiently saturated rendering of all colours acts as an animating element. Colour rendering in television technology is essential for a believable image transfer. The camera sees the world through slightly different weighting functions than the human eye, especially the colour blue. However, the footage played back should represent reality as faithfully as possible. The cornerstone here is again the rendering of skin colour and 'memory' colours. Each of these areas suits a different scale of colour rendering. A large number of candidates can be found in the literature, especially in [3], from which the author of this paper draws extensively. In practice, however, these quantities are not widely used [4]. Many of them, however, reveal the weaknesses of the algorithm currently used (see Figure 1) and suggest how to construct a new, more accurate measure that better captures the average subjective assessment of artificial light by a large number of people, for the whole spectrum of light sources with different parameters.

A Brief History of Ra (CRI)

The difference between the colour rendering of objects in sunlight or incandescent and fluorescent light has accompanied lighting technology since the 1930s. In 1937, the method of spectral deviations from the ideal source in eight bands was developed [5] and recommended by the CIE in 1948 as the color rendering measure (SBM). CIE Publication 13.1:1964 introduced the terminology and method of eight color samples. CIE 13.2:1974 introduced reference emitters, added six additional color samples (saturated colors, skin and plant colors), von Kries chromatic adaptation, and calculation in UCS coordinates (see Figure 1). CIE 13.3:1994 corrected several errors, added a sample implementation (see Figure 2), and is a valid method for determining the general color rendering index for light sources. Outside Europe, the designation CRI (Color Rendering Index) is often used for Ra. In our country, the CRI designation is mainly used by vendors, who have kept Ra a secret for the time being. Ra was originally designed for fluorescent lamps, and its calculation involves a constant that was set so that the Ra of the warm-tone halophosphate lamps of the time was approximately 50. However, with the advent of light-emitting diodes, this definition is no longer sufficient because people subjectively rate their light in terms of colour rendering better than the Ra value would correspond - in some cases by as much as ten points. Ra "punishes" colour shifts in all directions equally. Yet a shift towards a richer colour can be pleasant, and a small reduction in saturation while maintaining hue is much more acceptable than, for example, an equally large shift of yellow towards green. Alternative measures of colour rendering In the search for a better way to calculate the colour rendering index, many methods build on the algorithm for calculating Ra (see Figure 1) and trade off individual function blocks for more refined versions of them, e.g. a more appropriate colour space, a different set of colour samples, or a different averaging method. Other methods add transformations to the algorithm to improve it. Other methods are based on empirically determined computational procedures. An interesting group consists of experimental methods that evaluate light based on how well volunteers perform a color-related task under it, or how they subjectively evaluate the naturalness of color rendition of familiar objects.

R96a is one of the CIE's proposals for the Ra upgrade This proposal did not take off mainly due to disagreements between researchers and manufacturers (CIE 135-1999). It includes a new set of ten test samples: eight from the Color-Checker swatch and two skin colors, introduces new reference lights, a new chromatic adaptation, the calculation of color differences in CIELAB space, and the adaptation of all colors to D65.

Ra215 is the method proposed in [6]. This method is based on 215 colour samples, which include very saturated colours. The value is given, for example, by the manufacturer of calibrated fluorescent lamps for DTP Just Normlicht. A few years ago, Philips Lighting fluorescent and discharge lamp data sheets included diagrams showing the colour ring and shifts of these 215 samples in the a*, b* coordinates.

CRI CAM02UCS is an improvement of the Ra algorithm. This method uses 35 color samples, a newer color space and the CAM02-UCS color perception model [7]. Remarkably, despite all these improvements, it performs poorly in subjective tests, see [3].

CQS (Color Quality Scale) This proposal is being developed at the American Standards Institute (NIST). It is primarily aimed at improving the numerical performance of light emitting diodes, in line with how people evaluate light. It uses richer color samples, CIELAB color space, and CMCCAT2000 chromatic adaptation. The basic variant of CQSa (Qa) does not "punish" deviations towards more saturated colors, CQSp both "rewards and punishes" them, and CSQf is a measure of color fidelity [3 p. 8154].

FSI (Full Spectrum Index) The "Full Spectrum Index" expresses the difference between the spectrum of the source under investigation and the isoenergetic (full) spectrum (Tc = 5 455 K, Ra = 95). It is followed by the FSCI (Full Spectrum Color Index) - "full spectrum color index". It is rather a mathematical construction [8].

GA (Gamut Area) The "Gamut Area" is the area of an octagon in the coordinates u', v', whose vertices are the coordinates of the colour samples for Ra illuminated by the light under examination. A larger GA means a greater coverage of the colour space and therefore the inclusion of more saturated colours. GA increases with Tc. It is especially popular in Japan. A similar procedure is used by CSA (Cone Surface Area) [3 p. 8155-6].

GAI (Gamut Area Index) The "color space coverage index" is 100 times the ratio of the GA of the source under investigation to the GA of the isoenergetic spectrum [3 p. 8155-6]. The arithmetic average of Ra and GAI (denoted GAI_Ra) has been shown to be the best measure of the "naturalness" of light [3 p. 8166]. GAI expresses the rendition of saturated colors. Ra evaluates only the magnitude of color variation and not its direction, which just carries information about saturation change. Perhaps this is why the two measures complement each other so well.

RÖlfarben - "oil colour rendering index" This scale was proposed in the magazine Licht, Nos. 1-2 and 3/2013, for comparing light sources in terms of suitability for use in art museums. It is based on the Ra algorithm and uses 79 colour samples of oil and metallic paints from the painter's palette. The results are closely correlated with Ra, indicating that the set of color samples for Ra is very well chosen. A variation of Ra for rendering the colors of teeth and crowns of dental implants yielded similar results.

RCRI (Rank order Color Rendering Index) This "Rank order Color Rendering Index" is an experimental measure. Volunteers are asked to grade the color difference and several other variables between reference and test light for seventeen color samples. Based on the number of 1s and 2s in the overall rating, the RCRI value is determined. Thus, this method works directly with humans [3 p. 8155].

CCRI (Categorical Color Rendering Index) "Categorical Color Rendering Index" is a similar experimental approach proposed in Japan. The experiment involves classifying 292 color samples illuminated by the source under investigation into eleven word color categories. Based on these experiments, the boundaries of each named region in the CIECAM97 color space were determined. With their knowledge, the successful classification of samples under the investigated light can be predicted.

FCI (Feeling of Contrast color rendering Index) The "Feeling of Contrast color rendering index" is 100 times the ratio of the areas of two quadrilaterals in CIELAB space whose vertices correspond to samples of deep red, yellow, green and blue. The first quadrilateral consists of the samples illuminated by the light under study, the second of the samples illuminated by D65. Thus, it expresses the decrease in saturation of the primary colors compared to daylight [3 p. 8155].

(MCRI) Sa (Memory Color Quality metric) The "Memory Color Quality Metric" uses ten common color stimuli (apple, banana, orange, lavender, strawberry yogurt, cucumber, cauliflower, human skin, and a Smurf character). Using the distributions obtained in psychophysical experiments, the similarities of the ten displayed and memory (reference) colors are determined. Sa is the geometric mean of these similarities, and if it approaches one, all objects look as expected in the examined light [3 p. 8156].

Rf (Judd's Flattery Index) This is a 1967 proposal to improve Ra, when concerns arose that Ra did not reflect the public's preference for colour rendering. It uses ten color samples, eight similar to Ra, plus a sample of leaf green and human skin color. For each of these, a preferred colour shift is determined, by one-fifth of which the samples illuminated by the reference light are shifted. The procedure is based on the need to add a skin sample and to add 'directionality' (resolution of hue and saturation variation) to the Ra calculation. The CPI (Thornton's Color Preference Index), or "Thornton's Color Preference Index" [3 p. 8156-7], is based on a similar principle.

MIvis and MIuv - the MI metameric indices are used in the evaluation of light sources for daylight simulators for colorimetry, particularly in the graphic arts industry. (Related standards: ISO 3664, ISO 23603 and CIE 52.) The aim of these procedures is to avoid a situation where a colour difference that would be obvious in daylight cannot be distinguished in artificial light. The vis variant only deals with the visible area, while the uv variant also takes into account UV radiation, under which e.g. chlorine-bleached paper visibly luminesces.

TLCI (Television Lighting Consistency Index) The "Television Lighting Consistency Index" works with technical characteristics such as the spectral efficiency of camera colour sensors or gamma correction. Light-emitting diodes are not only making their way into reportage floodlights, but also into the lighting of television studios. TLCI allows to compare the different sources in terms of suitability for lighting in television camera shooting [9].

Ra,2012 (CIE CRI2012, nCRI) Ra,2012 is one of the most recently proposed methods. It uses seventeen artificial colors (bases) (see Figure 3), chosen based on the analysis of a catalog of 100,000 color spectra [10]. For Ra, manufacturers could optimize the submission of eight test samples at the expense of the other colors, which is actually excluded here. The method works with the CAM02-UCS color space, quadratic averaging of special rendering indices, and nonlinear weighting. The kind reader can find a list of programs for calculating various color rendering scales at the link http://goo.gl/nHjQ7R [cited October 31, 2014].

Comparison of colour rendering scales

Fourteen different color scales were compared in [3] (Ra,2012 was not yet available). Volunteers rated the light of different light sources in terms of attractiveness (appreciation) and naturalness (naturanlness). The different methods were ranked according to the success with which they predicted these ratings, see Table 1. A negative correlation was found between attractiveness and naturalness ratings. Thus, both subjective variables have their place, and it is possible that two numbers are needed for a more complete representation of color rendition.

Conclusion

It is impossible to capture the colour rendering for different purposes with one number. Different scales can be used to compare light sources for specific purposes. The current state of Ra has its bright side: LEDs shine "better" than Ra says they do. The lagging definition of Ra thus partly compensates for the need to progress in colour rendering quality requirements.

Literature:

[1] EN 12464-1 Light and lighting - Lighting of work areas: Part 1: Indoor work areas. March 2012. Prague, Office for Technical Standardization, Metrology and State Testing, 2012.

[2] CLAUSEN, H.: Light: nature as a reference in lighting design [on-line]. 1st ed. Meldorf, Hansen, 2009, ISBN 978-879-2154-026 [cited 2014 Oct 31]. Available from: http://goo.gl/v3RHl0.

[3] SMET, K. - RYCKAERT, W. R. - POINTER, M. R. - DECONINCK, G. - HANSELAER, P.: Correlation between color quality metric predictions and visual appreciation of light sources. Optics Express, 2011, vol. 19, issue 9, pp. 8151-8166. [cited 2014 Oct 31]. Available from http://goo.gl/ pskg4J.

[4] ŽÁK, P. - HABEL, J.: Influence of spectral composition of source radiation on observers' subjective perception in public lighting conditions. www.osvětle.cz [online]. 2014. [cited 31 October 2014]. Available from: http://goo.gl/aJ0tY6.

[5] HUNT, R.: Measuring Colour. 2nd Ed. New York, Ellis Horwood, 1991, 313 p., ISBN 01-356-7686-X.

[6] OPSTELTEN, J. J.: The establishment of a representative set of test colours for the specification of the colour rendering properties of light sources. CIE 20th session, 1983, D112/1-4.

[7] LI, Ch. - LUO, M. R. - LI, Ch. - CUI, G.: The CRI-CAM02UCS colour rendering index. Color Research, 2012, vol. 37, issue 3, pp. 160-167. [cited 2014 Oct 31]. Available from http://goo.gl/56iA40.

[8] What is full-spectrum index? Rensselaer Polytechnic Institute. Lighting Research Center [online]. 2004 [cited 2014 Oct 31]. Available from: http://goo.gl/fhb1Lj.

[9] Television Lighting Consistency Index 2012. EBU. EBU Technology & Innovation [online]. 2012. [cited 2014 Oct 31]. Available from: http://goo.gl/W73N18.

[10] KHANH, T. Q. - BODROGI, P. - VINH, T. Q. - BRÜCKNER, S.: Farbwiedergabe von konventionellen und Halbleiter-Lichtquellen: Theorie, Bewertung, Praxis. Munich, Pflaum, 2013, ISBN 978-379-0510-324.

[11] YAGUCHI, H.: Status quo of CIE work on colour rendering indices. 2013.

[12] HABEL, J.: Fundamentals of lighting technology (5). Světlo, 2009, No. 6, pp. 53-57. [cited 31.10.2014]. Available from: http://goo.gl/WcNDLP.

Review: doc. Ing. Michal Vik, Ph.D., Technical University of Liberec

Fig. 1. Algorithm of Ra calculation - see [12]

Table 1: Brief evaluation of the experiment comparing different scales

Table 2. Scales suitable for different purposes - according to [11]

Fig. 3. Spectral reflectances of the HL17 artificial colour set for Ra,2012 - data [11]

Fig. 4. CT&A CRI window

Author. Antonín Fuksa, NASLI & Blue step, spol. s r. o.
Published in Světlo 6/2014