Is the new CIE color
rendering index finally
in sight?
Dr. Kevin Smet
Light&Lighting Laboratory
Accelerating LED lighting, November 6, 2015
What is color?
What
is
Color
Rendering?
What
is
Color
Rendering?
What is color rendering?
A light source can induce different types of color distortions
b’
Hue shift
Reference
color
a’
Saturating
shift
Desaturating
shift
Equal sized color distortions but different color appearance!
color rendering
CIE Color rendering:
“Effect of an illuminant on the color appearance of objects by conscious or
subconscious comparison with their color appearance under a reference
illuminant.”
CIE Color Rendering Index, Ra
Test Sample Method, CIE13.3 (1995)
= objective measure for fidelity
color rendering
Reference
illuminant of
same CCT
Test lamp
Planckian radiator
CCT ≤ 5000 K
Daylight phase
CCT > 5000 K
color rendering
What CIE CRI conveys:
What CIE CRI doesn’t convey:
• Ra : average color fidelity/color
shift
• Ri : color fidelity/color shift for
specific hue region
• Direction/type of color shifts
• Information on human preference,
naturalness, discrimination, …
• Difference in color for any specific
object
(disregarding limitations of
samples and color science)
• How one source will make things look
compared to another
New color rendering index proposal
• CIE CRI is inaccurate at assessing color fidelity:
– Color science is outdated (color space, chromatic adaptation…)
 use up-to-date color science
– Set of test samples is imperfect
 use improved sample set
Color Fidelity Index, Rf
• CRI only tells us ONLY about color fidelity
 add a second number:
Color gamut measure Rg
 add a color distortion diagram
Proposal is based on various elements proposed over past years
Recommended and approved by Illuminating Engineering Society (IES): TM30-15
IES color rendition framework
Contributors:
Academia (K. Houser, Y. Ohno, M. Royer, K. Smet, M. Wei, L. Whitehead),
Industry (A. David, P. Fini), Lighting Design (R. Burkett)
CIE TC1-90
CIE TC1-91
Dual metric proposal (Rf, Rg) + lots of graphical info
100
95
90
Rf by hue
85
80
75
70
65
60
55
50
Color science update
J’
a’
• Replace outdated U*V*W*
with state-of-the-art color
space CAM02-UCS
 good perceptual uniformity
 no CCT dependence
 includes a good chromatic
adaptation formula
 includes a color difference formula
b’
Sample set improvement
• Replace CIE CRI Munsell test color
samples (TCS) with special Color
Evaluation Samples (CES):
1.
Larger sample size
J’
2.
Uniformly distributed (3D) in color space
a’
3.
Spectral or wavelength uniformity
b’
Sample set improvement
 Larger sample size
 from 8 (14) to 99 samples
100
95
90
Rf by hue
85
80
75
70
65
60
55
50
More information
Better statistical accuracy
Sample set improvement
 Uniform 3D distribution in color space
 Start from 105 000 reflectance samples
o
Natural objects, Paints, Plastics, Fabrics,
MacAdam limit
(all possible colors)
Printed materials, Skin tones…
NCS gamut
(common colors)
 Discard some colors:
o
Extremely saturated or dark
o
No color-error formula
b’
 Uniformly distributed selection
o
4880 color points
All color in database
a’
Wavelength uniformity
We need to make sure that samples treat all wavelengths equally.
Why? It is possible to generate many colors with only 3 “pigments”!
400
Reflectance
Pigment reflectance
100%
500
600
l (nm)
700
0%
400
500
l (nm)
600
700
But the corresponding samples are mostly sensitive to a few wavelengths
Wavelength uniformity
We can compute the “wavelength sensitivity” for a sample set (r’2, r”2…)
100%
-4
Reflectance
5
x 10
4.5
Sensitivity to SPD variations
4
0%
3.5
3
2.5
2
1.5
1
0.5
0
400
Black = 3-pigment set
Blue= TCS 1-8
Red = 4880 reference set
400
450
500
550
l (nm)
600
650
700
500
l (nm)
600
700
Sample set improvement
100
Each point = one SPD
a’
b’
criteria
90
Rf (99 test samples)
J’
80
70
60
50
50
60
70
80
90
100
Rf (5.000 test samples)
 99 Color Evaluation Samples
-4
1.5 x 10
Spectral sensitivity
Further downsampling
 4880 Reference set
J’
1
0.5
0
a’
b’
400
450
500
550
l (nm)
600
650
700
Sample set improvement
Impact of wavelength uniformity on Rf scores
11 sets of varying “wavelength uniformity” (N=2000)
Sample set improvement
Impact of wavelength uniformity on Rf scores
11 sets of varying “wavelength uniformity” (N=2000)
Set of 139 light sources
(broadband and narrowband
LED and fluorescent lamps)
Clear systematic effect
of degree of
“wavelength uniformity” !
Set of 31 triband fluorescent lamps
IES Rf versus CIE Ra
100
LER
Note: it is the spread that
matters, rather than the
correlation coefficient
80
400
350
Rf
60
40
300
20
250
0
0
20
40
60
Ra
80
100
200
???
Selective spectral
optimization
by taking advantage of
“wavelength
NON-uniformity”
of CIE test samples?!
Color Rendering Example
Comparison between existing (common) light sources having the same Ra but different Rf
light source 1
2700 K LED
Ra = 84, Rf = 83
light source 2
2700 K fluorescent
Ra = 82, Rf = 70
12 worst samples (LED
(fluorescent
source)source)
(test samples shown: 5, 8, 25, 29, 30, 31, 32, 33, 38, 48, 86, 94)
Ra = 84, Rf = 83
2700 K LED
Reference illuminant
2700 K fluorescent
Ra = 82, Rf = 70
Differences
in predictions
Ra and Rf
Here both LED
and FL arebetween
quite similar
are
readily visible
in this
“on-screen
 overall,
LED and
FL are
similar for some
experiment”:
some
samples
yellow
and
samples but
FL is
clearly(esp.
worse
for others
green) are visibly worse under the fluorescent
source.
Color Rendering Example
Comparison between an existing LED source and a possible narrowband
source, having the same Ra but different Rf
light source 1
light source 2
Ra = 81, Rf = 80
Ra = 80, Rf = 49
Hyperspectral images rendered with IES 4900 Refset under 3000 K
Ra = 100, Rf = 100
Ra = 100, Rf = 100
Hyperspectral images rendered with IES 4900 Refset under 3000 K
Ra = 81, Rf = 80
Ra = 80, Rf = 49
Reference illuminants
CIE CRI system had discontinuity at 5000 K because of abrupt switch from
Planckian radiators (P) to Daylight phases (D).
Proposed solution:
Gradual change by mixing P and D reference illuminants
Daylight locus
5000K
4500K
5500K
5000K
4500K
Blackbody locus
4000K
Summary
• CIE Ra has outdated color science and imperfect samples leading to
inaccurate assessment of color fidelity.
• The proposed Rf fixes this by:
o Updating color space to uniform CAM02-UCS
o Improving the color samples:
 Increased number provides more info and better statistical accuracy
 Spectral uniformity eliminates wavelength bias ensuring selective spectral optimization
becomes much harder.
o …
• Fidelity scores do not convey information on:
o
o
Preference, naturalness, discrimination, …
Direction and type of color shifts
o
…
• Extra info on color rendition properties can be provided by:
o
Graphical information, color gamut measure Rg or other (additional) measures such as
MCRI, CQS Qp, Harmony index, FCI, (see work of CIE TC1-91) …
Questions or comments?
Kevin.Smet@kuleuven.be
More info:
• Smet, Kevin A.G., David, Aurelien, & Whitehead, Lorne. (2015). Why color
space uniformity and sample set spectral uniformity are essential for color
rendering measures. Advanced online view LEUKOS.
doi: 10.1080/15502724.2015.1091356
• David, Aurelien, Fini, Paul T., Houser, Kevin W., Ohno, Yoshi, Royer,
Michael P., Smet, Kevin A. G., . . . Whitehead, Lorne. (2015). Development
of the IES method for evaluating the color rendition of light sources. Optics
Express, 23(12), 15888-15906.
doi: 10.1364/OE.23.015888
• IES. (2015). IES-TM-30-15: Method for Evaluating Light Source Color
Rendition (pp. 26). New York, NY: The Illuminating Engineering Society of
North America.
Thanks to all collaborators on this project:
Aurelien David, Paul Fini, Kevin Houser, Yoshi Ohno, Michael Royer, Minchen Wei, Lorne Whitehead