An alternative colour rendering index based on memory colours Kevin SMET KaHo Sint-Lieven – K.U.Leuven Light & Lighting Laboratory – ESAT/ELECTA Gebroeders Desmetstraat 1, 9000 Gent (Belgium) Tel: +32 92 65 87 13 kevin.smet@kahosl.be Kevin Smet – An alternative colour rendering index

All data is for personal or TC1-69 use. The basic idea, methods used, visual results & modeling have been written in a paper that is under review for publication in “Journal of Vision”. Results comparing the Memory Colour based index with CIE CRI, CQS and visual results like the Thurstone scalings, kindly provided by Sophie Jost-Boissard, as well as with visual scaling experiments that will be performed in the near future will be published in a next paper. Kevin Smet – An alternative colour rendering index

Basic Idea & Theoretical Method Kevin Smet – An alternative colour rendering index

Basic Idea: Absolute assessment No need of reference illuminants! No dependence on correlated colour temperature: test sources with different CCT can be compared. More realistic way of evaluating colour rendering. (*) Memory colours define the colours that are recalled in association with familiar objects in long-term memory (Bartleson, 1960).. Referencing is done to memory colours* of familiar objects, NOT to a reference illuminant! Kevin Smet – An alternative colour rendering index

Method Investigate colour acceptance boundaries of familiar objects Test subjects rate acceptability (YES/NO) of a set of apparent object colours: YES / NO Only a qualitative measure for colour rendering based on acceptability of apparent object colour. Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 5

Method Test subjects rate similarity (on a 5 point scale) of a set of apparent object colours to their memory colour of the object: Similarity ratings on a 5 point scale as a function of chromatic distance to memory colour. Quantitative measure for colour rendering. Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 6

From similarity ratings to an alternative colour rendering index Model pooled similarity ratings by a bivariate Gaussian surface R(x,y): Chromaticity of apparent object colour: X Centre Xc = chromaticity of memory colour Mahalanobis distance d(x,y) ~ degree of similarity S(x,y) ai are fitting parameters: a1 & a2 are used to account for inter observer variability a3 to a4 describe the similarity function S(x,y). Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 7

From similarity ratings to an alternative colour rendering index Alternative colour rendering index Sa: Calculate for each object i: chromaticity of object rendered by the test source by using the spectral reflectance factors of the object. Mahalanobis distance di to chromaticity of memory colour Special colour rendering index: S’i = exp(-0.5* di) General colour rendering index: S’a= mean(Si). 5. (Rescale indices to CRI levels: Ra(F1-F12)= a*Sa(F1-F12)+b) Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 8

Practical Kevin Smet – An alternative colour rendering index

Real versus simulated objects? Naturalness has influence on similarity ratings (Yendrikhovskii, Blommaert & De Ridder, 1999) Colour constancy is greater for surface colours and lights in real scenes than in simulated scenes. (Berns & Gorzynski, 1991; Brainard, 1998) Experiments with real objects ! Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 10

Illumination box to change apparent object colours Special design to mask any illuminant clues! a. RGBA LEDs change apparent object colour b. Diffusing tunnel: uniform illumination of objects avoid specular reflection on object c. Transparent support for objects: hide surface reflections revealing illumination colour d. Back panel lighting: constant adaptation white with CCT = 6000K. Create the illusion of the object changing colour! Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 11

Spectral Measurements Tele-spectroradiometrically: radiometric measuring head 74055 MS260i spectrograph iDUS DV420A-BU2 CCD camera Setup identical to viewing conditions Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 12

Chromaticities / colour spaces CIE 10 degree observer Measured chromaticities are transformed to corresponding colours: Chromatic Adaptation Transform: CAT02 Equal Energy Illuminant Colour spaces: SVF (Seim & Valberg, 1986): ~ perceptually uniform colour space CIECAM02 is unfortunately not applicable: Yb of background cannot be defined, because it is more luminous (=self luminous) than the reflected white of the front panel (Yb would be > 100). Personal communication with Mark Fairchild. Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 13

Test Object Selection green apple, ripe banana, … High degree of familiarity: NATURAL OBJECTS (e.g. food) Object chromaticity distributions are preferably unimodal or a single mode should be unambiguously specifiable; i.e. the natural colour variance should not be too high: e.g.: green apple, ripe banana, … but NOT: blue grape, because the chromaticity of blue grapes varies from a black/dark blue to wine red. Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 14

Test Object Selection Object chromaticities spread uniformly around the hue circle: Primary regions of choice: green, yellow, red, purple & blue. BUT some difficulties: Illumination box cannot change the apparent colour of RED natural objects sufficiently due to their highly chromatic nature. - ORANGE is taken as a compromise. Number of blue well-known unambiguously specifiable objects is extremely limited. - a SMURF figurine was selected, because when questioned most people report having a good idea smurf blue. Extra experiments are planned with some less chromatic objects as well: strawberry yoghurt (red), cheese , caucascian skin, … Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 15

Experiment setup: Calibrate illumination box: map XYZ to RGBA DAC input values. Calculate object colour solid. Select luminance plane with optimum chromaticity gamut. Calculate RGBA DAC input values resulting in a uniform grid in the colour space selected. Add double grid points to illumination sequence. Randomize sequence. Add a set of 20 training points to familiarize observers with the scale. Measure the object spectral radiance. Calculate chromaticities of sequence of apparent object colours Visual assessment of sequence by test subjects: Similarity ratings on a 5 point scale 1: very bad; 2: bad; 3: neutral; 4: good; 5: very good Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 16

Experimental results Kevin Smet – An alternative colour rendering index

Visual assessment results Intra & Inter observer variability with PF/3 performance factor: Results are good, considering that a PF/3 of 30 is common in colour discrimination studies. (Guan & Luo, 1999a,b; Xu, Yaguchi & Shiori, 2001) Data can be pooled and average observer can be postulated. Apple Banana Orange Lavender Smurf Intra PF/3 19 24 30 Inter PF/3 39 31 37 45 41 Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 18

Visual assessment results Average observer reliability (real world validity) analysed with ICC(2,n) according to method described by Shrout & Fleiss (1979). Good results: validate the use of pooled rating data as basis for colour rendering evaluation. ICC(2,n) shows a high inverse correlation (ρ = -0.94) with the PF/3. Observers agreed best on what a ripe banana looks like and least on what dried lavender should look like. Apple Banana Orange Lavender Smurf ICC(2,n) 0.94 0.97 0.92 0.86 0.91 Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 19

Model results Pooled rating distribution was fitted using MATLAB lsqcurvefit function. Goodness-of-fit to mean ratings of pooled data analysed with Pearson ρ and RMSE: Confirmation of Yendrikhovskij et al. (1999): Similarity judgements can be described by a bivariate Gaussian distribution in a perceptually uniform colour space. Apple Banana Orange Lavender Smurf ρ 0.95 0.97 0.94 0.96 RMSE 0.28 0.21 0.30 0.24 0.23 Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 20

Colour rendering evaluation based on memory colours. - Results - Kevin Smet – An alternative colour rendering index

Rescaling the similarity functions Similarity functions are rescaled to the CIE CRI level by a linear transformation calculated using a least squares approach: Advantage: Colour rendering scores for old sources are kept nearly the same. Fluorescent sources F1-F12 Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 22

Alternative CRI for 90 lamps Lamp spectral radiances: see excel file. Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 23

Correlation with visual assessments All visual scaling data was kindly provided by Sophie Jots-Boissard The Pearson correlation coefficient between the Thurstone scalings for attractiveness, of nine 3000K (Boissard, 2009) & eight 4000K lamps obtained in a series of visual experiments performed by Sophie Jost-Boissard, and the Memory Colour based index is high: ρ3000K = 0.95 (p<0.05) & ρ4000K = 0.87 (p<0.05). As to not obscure any possible correlation between the visual assessments and the Memory Colour based index, only the special colour rendering indices (apple, banana and orange) have been used in this calculation, since a light source having terrible colour rendering capabilities in the blue to purple region might obscur this correlation. The Pearson correlation has also been calculated between the Thurstone scalings for attractiveness and the CRI and CQS. The Pearson correlation coefficients were: ρCRI,3000K = 0.26 (p=0.50) & ρCRI,4000K = 0.07 (p=0.86) ρCQS,3000K = 0.48 (p=0.20) & ρCQS,4000K = 0.44 (p=0.28) For the same reason descibed above, special colour rendering indices 2,3,4,10,11,13 & 14 have been used for the CRI, while for the CQS special indices 7-13 were selected. Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 24

Planned experiments Extend special indices into the red region: strawberry yoghurt, ham, bacon,… Add some less saturated objects. Try and get a similarity function for Caucasian skin. Extend number of observers on each object. Perform validation experiments with a set of test lamps having low CRI but high visual appreciation and with an object set that is not limited to the green-red region! Change to an even more perceptually uniform colour space: eg. IPT? Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 25

References Kevin.Smet@kahosl.be 26 Bartleson, C. J. (1960). Memory colors of familiar objects. J. Opt. Soc. Am., 50, 73–77.   Berns, R.S. & Gorzynski, M.E. (1991). Simulating surface colors on CRT displays: the importance of cognitive clues. Proc. AIC Colour and Light, 91, 21-24. Brainard D.H., (1998). Color constancy in the nearly natural image. J. Opt. Soc. Am. A, 15, 307-325. Guan, S. & Luo, M.R. (1999a). Investigation of parametric effects using small colour differences pairs. Col. Res. Appl., 24, 331–343. Guan, S. & Luo, M.R. (1999b). A colour-difference formula for assessing large colour-differences. Col. Res. Appl., 24, 344-355. Jost-Boissard, S., Fontoynont, M., & Blanc-Gonnet, M., (2009). COLOUR RENDERING OF LED SOURCES: VISUAL EXPERIMENT ON DIFFERENCE, FIDELITY AND PREFERENCE. Proc. CIE Light and Lighting conference. Budapest. Hungary. 27-29 may 2009.     Seim, T. & Valberg, A. (1986). Towards a uniform color space. A better formula to describe the Munsell and OSA color scales. Col Res. Appl., 11, 11–24. Xu, H., Yaguchi, H. & Shioiri S. (2001). Estimation of Color-Difference Formulae at Color Discrimination Threshold Using CRT-Generated Stimuli. Optical Review, 8 (2), 142-147. Yendrikhovskij, S. N., Blommaert, F. J. J., De Ridder, H. (1999). Representation of memory prototype for an object colour. Col. Res. Appl., 24 (6), 393–410. Kevin Smet – An alternative colour rendering index Kevin.Smet@kahosl.be 26

Kevin Smet – An alternative colour rendering index