SIGGRAPH 2016
Optimal LED Selection for Multispectral Lighting Reproduction
Y
7.00E-03
White
6.00E-03
Royal Blue
Blue
5.00E-03
B
4.00E-03
Green
C
G
W
1.00E-03
450
500
Amber
A
L
Red Orange
Red
P
550
Deep Red
600
650
700
Figure 1: Custom circuit board with 11 different Lumileds Luxeon Rebel ES LEDs, with measured emission spectra.
Abstract
We demonstrate the sufficiency of using as few as five LEDs of distinct spectra for multispectral lighting
reproduction and solve for the optimal set of five from 11 such commercially available LEDs. We leverage
published spectral reflectance, illuminant, and camera spectral sensitivity datasets to show that two approaches
of lighting reproduction, matching illuminant spectra directly and matching material color appearance observed
by one or more cameras or a human observer, yield the same LED selections. Our proposed optimal set of five
LEDs includes Red, Green, and Blue with narrow emission spectra, and broader-spectrum White and PC Amber.
6 LED RGBW + LP
6 LED RGBW + LA
0.03
6 LED RGBW + PA
5 LED RGBW + A
5 LED RGBW + P
0.02
5 LED RGBW + L
4 LED RGBW
0.01
3 LED RGB
Approaches
Datasets
• Spectral Illuminant Matching (SIM)
• Illuminants - Direct
• Reflective Materials
F4
Skylight
D65 x Grass
chart
6 LED RGBW + LP
3.00%
6 LED RGBW + LA
2.50%
6 LED RGBW + PA
5 LED RGBW + A
2.00%
5 LED RGBW + P
1.50%
5 LED RGBW + L
4 LED RGBW
1.00%
3 LED RGB
0.50%
D65 x Sand
A
D65
F4
Skylight
D65 x Grass
DEBEVEC, P., WENGER, A., TCHOU, C., GARDNER, A., WAESE, J.,
AND HAWKINS, T. 2002. A LIGHTING REPRODUCTION APPROACH TO
LIVE-ACTION COMPOSITING. IN PROC. 29TH ANNUAL CONFERENCE ON COMPUTER
GRAPHICS AND INTERACTIVE TECHNIQUES, ACM, NEW YORK, NY, USA,
SIGGRAPH ’02, 547–556.
JIANG, J., LIU, D., GU, J., AND SUSSTRUNK, S. 2013. WHAT IS THE SPACE
OF SPECTRAL SENSITIVITY FUNCTIONS FOR DIGITAL COLOR CAMERAS. IN APPLICATIONS
OF COMPUTER VISION WACV, 2013 IEEE WORKSHOP ON, 168–179.
LEGENDRE, C., YU, X., LIU, D., BUSCH, J., JONES, A., PATTANAIK,
S., AND DEBEVEC, P. 2016. PRACTICAL MULTISPECTRAL LIGHTING
REPRODUCTION. ACM TRANS. GRAPH. 35, 4 (JULY).
PROCEEDINGS OF THE 14TH EUROGRAPHICS WORKSHOP ON
RENDERING, EUROGRAPHICS ASSOCIATION, AIRE-LA-VILLE, SWITZERLAND, EGRW
’03, 249–259.
WENGER, A., HAWKINS, T., AND DEBEVEC, P. 2003. OPTIMIZING COLOR
MATCHING IN A LIGHTING REPRODUCTION SYSTEM FOR COMPLEX SUBJECT AND
ILLUMINANT SPECTRA. IN
11 LED (One Observer Solve)
11 LED (All Observers Solve)
10.00%
11 LED (SIM Solve)
5 LED RGBW + P (One Observer)
8.00%
5 LED RGBW + P (All Observers)
5 LED RGBW + P (SIM Solve)
6.00%
4 LED RGBW (One Observer Solve)
4.00%
4 LED RGBW (All Observers Solve)
4 LED RGBW (SIM Solve)
2.00%
0.00%
D65 x Sand
A
D65
F4
Skylight
D65 x Grass D65 x Sand
color chart squares, for 29 observers, with
optimal LED weights computed using SIM,
separate MRM for each observer, or MRM for
all observers at once.
Illuminant A (MRM Human Observer)
Illuminant F4 (MRM Human Observer)
Illuminant D65 x Grass (MRM Human Observer)
0.3
0.3
11 LED
11 LED
0.25
5 LED (RGBWP)
0.2
5 LED (RGBWP)
3 LED (RGB)
0.15
0.1
0.2
4 LED (RGBW)
3 LED (RGB)
ACTUAL
0.15
0.1
0.05
400
450
500
5 LED (RGBWP)
4 LED (RGBW)
ACTUAL
0
11 LED
0.25
550
Wavelength (nm)
600
650
700
0
0.2
3 LED (RGB)
ACTUAL
0.15
0.1
0.05
400
450
500
550
Wavelength (nm)
600
650
700
0
400
450
500
550
Wavelength (nm)
600
650
700
Figure 7: Ground truth spectra (solid black lines) for illuminants A (left), F4 (center), and D65 modulated by grass (right). LED-reproduced
illumination is computed using Metameric Reflectance Matching for the CIE 2˚ standard observer for all 11 LEDs, RGBWP, RGBW, and RGB.
Conclusion
References
12.00%
color chart squares, for 29 observers, with
optimal LED weights computed separately for
each observer. Using more than five LEDs of
distinct spectra yields diminishing returns.
4 LED (RGBW)
Spectral sensitivities, 28 cameras [Jiang et al. 2013]
CIE 1931 2˚ standard observer
24 squares of the X-Rite ColorChecker
D65
0.05
TM
6 LED RGBW + CP
between reproduced and target illuminant
spectra, 400-700 nm, using Spectral Illuminant
Matching. Using more than five LEDs of distinct
spectra yields diminishing returns.
0.25
CIE D65 modulated by grass [USGS SPLIB06a]
CIE D65 modulated by sand (quartz)
MIM is under-constrained for more than three LEDs of
distinct spectra. We enforce MIM constraints in MRM,
using the flat spectra of neutral color chart squares.
A
0.3
• Illuminants - Indirect
• Cameras/Observers
3.50%
14.00%
Figure 4: Sum of Squared Errors (SSE) Figure 5: Average MRM color error for 24 Figure 6: Average MRM color error for 24
CIE A (incandescent tungsten lighting)
CIE D65 (average daylight)
CIE F4 (fluorescent lighting, CRI = 51)
Skylight (measured)
We produce LED light that yields color matches for
materials with known reflectance spectra, such as the
squares of a color chart, as seen by an observer.
11 LED
0.00%
0.00
Figure 2: Subject inside a multispectral light stage.
4.00%
Average error relative to white chart patch
intensity
6 LED RGBW + CP
0.04
Average error relative to white chart patch
intensity
11 LED
Neither [Wenger et al. 2003; LeGendre et al. 2016]
discussed a minimal, sufficient set of multispectral
LEDs for lighting reproduction, which we present.
• Metameric Reflectance Matching (MRM)
4.50%
0.05
LeGendre et al. [2016] reproduced omnidirectional
lighting in a 6-channel multispectral light stage [Fig. 2],
achieving improved color rendition using RGBW
lighting and very close color matches with RGBCAW.
We produce LED light that is metameric to a target
illuminant, to a given tristimulus observer with known
spectral sensitivities.
D65 x Sand
Metameric Reflectance Matching Error
Metameric Reflectance Matching Error
Spectral Illuminant Matching Error
Relative Power
Wenger et al. [2003] observed poor color rendition
using an RGB light source and showed improvement
when using a 9-channel multispectral light.
• Metameric Illuminant Matching (MIM)
D65 x Grass
observer, with illumination reproduced using various LED combinations. The background squares represent the ground truth computed color chart
appearances under the target illuminants, while the foreground circles represent the computed color chart appearances under LED-reproduced
illumination. Row 1 shows lighting reproduction results using all 11 LEDs of distinct spectra, row 2 shows results using 5 LEDs only (RGBWP), row
3 shows results using RGBW, and row 4 shows the result using RGB only. The RGBW lighting produces accurate color rendition for D65, Skylight,
and D65 modulated by grass and sand reflectance spectra. Adding PC Amber yields accurate color rendition for illuminants A and F4.
Lighting reproduction [Debevec et al. 2002] surrounds
a subject with Red, Green, and Blue (RGB) LEDs,
driving them to match the colors and intensities of a
real-world lighting environment, illuminating the
subject to appear as they would in the actual scene.
We produce LED light to match an illuminant spectrum.
Skylight
Figure 3: Color matching results using Metameric Reflectance Matching for various direct and indirect illuminants for the CIE 2° standard human
SSE Visible Spectrum
USC Institute for Creative Technologies
Introduction
Illuminant F4
Relative Power
400
PC Amber
Relative Power
2.00E-03
Lime
D
O
3.00E-03
0.00E+00
Cyan
R
Illuminant D65
RGBW LEDs RGBWP LEDs All 11 LEDs
Illuminant A
RGB LEDs
8.00E-03
Results
Material color appearance under various direct and indirect illuminants may be
accurately matched using as few as five LEDs of distinct spectra for multispectral
lighting reproduction: Red, Green, Blue, and broader-spectrum White and PC Amber.
Using more than these five LEDs yields diminishing returns.
Direct spectral illuminant matching and color appearance matching yield the same
optimal set of five LEDs for the illuminants and observers considered.
Acknowledgements
The authors thank Randy Hill, Kathleen Haase, Christina Trejo, and Jay Busch for their
important support of this work. This project was sponsored by the U.S. Army Research
Laboratory (ARL) under contract W911NF-14-D-0005 and in part by a USC Annenberg Ph.D.
Fellowship. The content of the information does not necessarily reflect the position or the
policy of the Government, and no official endorsement should be inferred.
Chloe LeGendre, Xueming Yu, Paul Debevec
USC Institute for Creative Technologies
http://gl.ict.usc.edu