Multimedia ImagingColor Science inChallenges forROY S. BERNS Chester F. Carlson Center for Imaging...
Transcript of Multimedia ImagingColor Science inChallenges forROY S. BERNS Chester F. Carlson Center for Imaging...
ROY S. BERNS
Chester F. Carlson Center for Imaging Science
Rochester Institute of Technology54 Lomb Memorial Drive
Rochester, New York [email protected]
http:/www.cis.rit.edu/people/faculty/berns/
ROY S. BERNS
Chester F. Carlson Center for Imaging Science
Rochester Institute of Technology54 Lomb Memorial Drive
Rochester, New York [email protected]
http:/www.cis.rit.edu/people/faculty/berns/
Challenges forColor Science in
Multimedia Imaging
Challenges forColor Science in
Multimedia Imaging
World Wide Web ChallengeThe Museum of Modern Art , New YorkWorld Wide Web ChallengeThe Museum of Modern Art , New York
Vincent van Gogh, The Starry NightVincent van Gogh, The Starry Night
ImagesImages
TextText
AudioAudio
J. Paul Getty Museum, Los Angeles, CAJ. Paul Getty Museum, Los Angeles, CA
Tone ReproductionTone Reproduction
Typical "Mac" SystemTypical "Mac" System Typical "PC" SystemTypical "PC" System
Color BalanceColor Balance
Typical "Mac" SystemTypical "Mac" System Typical "PC" SystemTypical "PC" System
Tone ReproductionTone Reproduction
Attack on Bunker Hill with the Burning of Charlestown(Unknown artist, circa 1783)Attack on Bunker Hill with the Burning of Charlestown(Unknown artist, circa 1783)
Color Management ChallengeColor Management Challenge
Color Managementvia Profile Connection Space (PCS)Color Managementvia Profile Connection Space (PCS)
drdgdb
CameraColorimetry
XYZ
Color AppearanceModel - CAM
(Scene conditions)
LCH
Inverse CAM(PCS conditions)
L*a*b*0
20
40
60
80
100
L*
-100-50050
100 a*
-100 -50 0 50 100b*
0
20
40
60
80
00
-100-50050
100 a*
drdgdb
LCH
CAM(PCS conditions)
Inverse CAM(Display conditions)
L*a*b*0
20
40
60
80
100
L*
-100-50050
100 a*
-100 -50 0 50 100b*
0
20
40
60
80
00
-100-50050
100 a*
GamutMapping
L*a*b*
Inverse CRTColorimetry
CameraProfile
CRTProflie
PCS
Images as Knowledge ChallengeImages as Knowledge ChallengeSeduction - Color PreferenceIntegrity - Color AccuracyLongevity - Image/Data ArchivesData Mining - Academic Pursuits
Seduction - Color PreferenceIntegrity - Color AccuracyLongevity - Image/Data ArchivesData Mining - Academic Pursuits
Seduction - Georgia O'KeeffeSeduction - Georgia O'Keeffe
+a*+a*
+b*+b*
-b*-b*
-a*-a*
CIELAB Color Gamuts:CRT vs. Photographic PrintCIELAB Color Gamuts:CRT vs. Photographic Print
Gamut "Mapping"Gamut "Mapping"
Color IntegrityTrichromatic AssumptionColor IntegrityTrichromatic Assumption
• Stimuli with the same specification match.• Stimuli that match have the same specification.• Stimuli with the same specification match.• Stimuli that match have the same specification.
X
Y
Z
1
=X
Y
Z
2
11 22
Assume same viewing conditionsAssume same viewing conditions
Multimedia Is Inherently MetamericMultimedia Is Inherently Metameric
7006005004000
20
40
60
80
100
Wavelength (nm)
Rel
ativ
e R
adia
nt P
ower
Color CRT
Caucasian Face in Daylight
The World Is Not Made From: Phosphors and InksThe World Is Not Made From: Phosphors and Inks
Printed PaperPrinted PaperCRT DisplayCRT Display
metsmets
show grum obs met slidesshow grum obs met slides
Inter-observer variance
Inter-observer variance
Intra-observer variance
Intra-observer variance
CRT and Photographic Media Metameric MatchingCRT and Photographic Media Metameric Matching
Alfvin and Fairchild, 1996.Alfvin and Fairchild, 1996.
CIE Standard Deviate ObserverCIE Standard Deviate Observer
0 10
∆ a*
-10
0
10
∆ b*
NimeroffNimeroff
CIE Standard Deviate ObserverCIE Standard Deviate ObserverIntra-ObserverIntra-Observer
Inter-ObserverInter-Observer
Color Measurement1931 Hardy Recording SpectrophotometerColor Measurement1931 Hardy Recording Spectrophotometer
Color MeasurementGretag SpectoScanColor MeasurementGretag SpectoScan
Colorant FormulationColorant Formulation
7006005004000.0
0.1
0.2
0.3
0.4
Standard
Metameric Match
Simple Color Difference
Wavelength (nm)
Re
fle
cta
nce
fa
cto
r
Metameric match has low inherent qualit yMetameric match has low inherent qualit y
Metameric Matches Are Unstable With Changes in LightingMetameric Matches Are Unstable With Changes in Lighting
6806205605004403803800.0
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)R
efle
ctan
ce
Fac
tor
6806205605004403803800.0
0.1
0.2
0.3
0.4
0.5
Wavelength (nm)
Ref
lect
ance
F
acto
r
Two painted surfaces, each colored with different pigments
Two painted surfaces, each colored with different pigments
Daylight IlluminationDaylight Illumination
What a typical observer sees:What a typical observer sees:
Incandescent IlluminationIncandescent Illumination
What a typical observer sees:What a typical observer sees:
The Camera Is Not A HumanThe Camera Is Not A Human
What the camera records under daylight:What the camera records under daylight:
The Camera Is Not A HumanThe Camera Is Not A Human
What the camera records under incandescent:What the camera records under incandescent:
Conventional Imaging Input Devices Are Not Linearly Transformable To Humans
Conventional Imaging Input Devices Are Not Linearly Transformable To Humans
7006506005505004504000.0
0.2
0.4
0.6
0.8
1.0
1.2
nm
Rel
ativ
e se
nsi
tiv
ity
≠≠
Two Issues: Analysis and SynthesisTwo Issues: Analysis and Synthesis
Printing systems are mainly four color, leading to metameric matches
Printing systems are mainly four color, leading to metameric matches
Digital input is notcolorimetricDigital input is notcolorimetric
Photographic input is not colorimetricPhotographic input is not colorimetric
Equivalent to...Equivalent to...
"Colorimeter""Colorimeter" Metameric MatchMetameric Match
The Spectral Challenge: The Spectral Challenge:
7006005004000.0
0.1
0.2
0.3
0.4
Standard
Metameric Match
Simple Color Difference
Wavelength (nm)
Re
fle
cta
nce
fa
cto
r
OriginalOriginal
ReproductionReproduction
A Solution: Multispectral-Based Color ReproductionA Solution: Multispectral-Based Color Reproduction
Image capture
Multi-channel image storage
Spectral-based printing
separation minimizing
metamerism
Multi-ink direct digital
printing
Copyright © 1993, The National Gallery, LondonCopyright © 1993, The National Gallery, London
Spectral reconstruction
Ink selection
Current Research EffortsMy ChallengeCurrent Research EffortsMy Challenge
Multispectral image capture
Spectral models of halftone printing
Statistical representation of paintings
Printing models minimizing metamerism
Color tolerances and spaces
Multispectral image capture
Spectral models of halftone printing
Statistical representation of paintings
Printing models minimizing metamerism
Color tolerances and spaces
Peter BurnsPeter BurnsPh.D. graduate, 1997Ph.D. graduate, 1997
Seven-Channel ColorimeterSeven-Channel Colorimeter
7006005004000.0
0.2
0.4
0.6
0.8
1.0
1.2
Wavelength (nm)
No
rma
lize
d
Tra
nsm
itta
nce
700600500400-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
Wavelength (nm)
no
rma
lize
d
cmf*
sou
rce
Thin lines = color matching functionsThick lines = least squares fit
Illuminant A weighted
Matrix TransformationL*a*b*L*a*b*
Define Dataset and Measure a Subset of the ImageDefine Dataset and Measure a Subset of the Image
Spectroradiometry or spectrophotometrySpectroradiometry or spectrophotometry
Describe the Spectral Properties: Principal Component AnalysisDescribe the Spectral Properties: Principal Component Analysis
Seven Filters + Digital CameraSeven Filters + Digital CameraImage sequentially followed by spectral estimationImage sequentially followed by spectral estimation
L*a*b*L*a*b*
Limitation:Digital cameras are low resolutionLimitation:Digital cameras are low resolution
2036x30602036x30603072x40963072x4096
VASARI: Visual Arts System for Archiving and Retreival of Images
VASARI: Visual Arts System for Archiving and Retreival of Images
National Gallery, LondonNational Gallery, London Uffizi Gallery, Florence, ItalyUffizi Gallery, Florence, Italy
Dr. Francisco ImaiDr. Francisco Imai
Postdoctoral FellowPostdoctoral Fellow
Combine:Low Resolution MultichannelHigh Resolution Luminance
Combine:Low Resolution MultichannelHigh Resolution Luminance
7006005004000.0
0.2
0.4
0.6
0.8
1.0
1.2
Wavelength (nm)
No
rma
lize
d
Tra
nsm
itta
nce
Compressing Color Channels Compressing Color Channels
After 16:1After 16:1
BeforeBefore
After 4:1After 4:1
Di-Yuan TzengDi-Yuan Tzeng
Ph.D. Candidate in Imaging SciencePh.D. Candidate in Imaging Science
Statistically Estimate PigmentsStatistically Estimate Pigments
Avg. MI=0.1 ∆E*94
Canonical Correlation AnalysisCanonical Correlation Analysis
The Optimal Ink- S
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
400 420 440 460 480 500 520 540 560 580 600 620 640 660 680 700Wavelength
R fl
Warm red
Purple
Reflex blue
Process blue
process black
process yellow
Koichi IinoKoichi IinoVisiting Scientist, ToppanVisiting Scientist, Toppan
Spectral Models of Color PrintingSpectral Models of Color Printing
R a R a R a R a Rc c
n
m m
n
cmyk cmyk
n
w w
n n
λ λλ
λλ
λλ
λλ λ= + + + +( ), , , ,...1 1 1 1
Optical Ink Interaction ModelOptical Ink Interaction Model
q f di i j t jj i
=≠
∏ _ ,( )
Results Estimating MatchprintResults Estimating Matchprint
325 independent colors sampling gamut 325 independent colors sampling gamut
Color Quality MetricsColor Quality MetricsThe Birth of CIELAB, Billmeyer, 1973The Birth of CIELAB, Billmeyer, 1973
City University, LondonCity University, London
WyszeckiRobertson
Ganz
MacAdam
Color Tolerance EquationsColor Tolerance Equations
∆ECH* = ∆L*
kLSL
2
+ ∆C*
kCSC
2
+ ∆H*
kHSH
2
1/2
where
SL = 1.0 SC = 1.0 + 0.045Cab* SH = 1.0 + 0.015Cab
*
9494
Hue Angle DependencyHue Angle Dependency
40030020010000.0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Luo-RiggRIT-Dupont
Witt
h
Del
ta
Hue
/(1+
0.01
5C*s
td)
0
1
2
3
4
-2 48 98 148 198 248 298
Hue angl e
T5
0
L60c20
L40c35
L40c20 red, L*=42.6, C*=42.6
greenL*=55.2, C*=33.7 blue
L*=35.0, c*=32.8
Previous DataPrevious Data Recent DataRecent Dataabab
Comparisons:CMC, BFD, Current ResultsComparisons:CMC, BFD, Current Results
0
0.5
1
1.5
2
2.5
-2 38 78 118 158 198 238 278 318
hue angle
The
ave
rage
hue
dis
crim
inat
ion
CMC BFDL*=40, C* ab=20 0.39 0.35L*=40, C* ab=35 0.62 0.67L*=60, C* ab=20 0.30 0.29Weightedaverage
0.45 0.45
RMS ∆H*ab errorRMS ∆H*ab error
CMC BFD
Hue LinearityHue LinearityHung and Berns, 1995.Hung and Berns, 1995.
150
50
100
0
-50
-100100500-50-100-150
(a) CIELAB space
(b) CIELUV space
150
50
100
-50
-100
-150
0
150100500-50-100-150 200
(d) Nayatani's model
a*b*
u*
v*
100
50
0
-50
-100
-1500 50 100-50-100
P
T
(c) Hunt's model
100500-50-100-150 150
0
-50
-100
-150
50
100
Myb
Mrg
Gus Braun and Fritz EbnerHue Corrected CIELABGus Braun and Fritz EbnerHue Corrected CIELAB
OriginalOriginalCIELABCIELAB Modified CIELABModified CIELAB
CIE TC 1-47CIE TC 1-47
Hue Angle Correction
Parametric Lightness Function
Evaluate Color Appearance Models
Consider New Color SpaceChroma Compression
Hue Linearity
Hue Angle Correction
Parametric Lightness Function
Evaluate Color Appearance Models
Consider New Color SpaceChroma Compression
Hue Linearity
Spatial + Color DifferenceSpatial + Color Difference
AchromaticRed-Green
Yellow-Blue
∆ ∆ ∆ ∆E
Lk S
Ck S
Hk SL L
ab
C C
ab
H H?*
* * */
=
+
+
2 2 2 1 2
Summary of ChallengesSummary of Challenges
Limitations of Colorimetric Matching
Practical Color Management
Practical Color Appearance Models
Spectral Color Reproduction
Remember the Basics
Limitations of Colorimetric Matching
Practical Color Management
Practical Color Appearance Models
Spectral Color Reproduction
Remember the Basics
Max Saltzman'sThree PrinciplesMax Saltzman'sThree Principles
Color is what is seen: light source, object, observer. Change one, change color.
The sample being judged must be representative of the entire batch of material.
Assess uncertainty in each step of a process.
Color is what is seen: light source, object, observer. Change one, change color.
The sample being judged must be representative of the entire batch of material.
Assess uncertainty in each step of a process.
"Color reproduction is a bloody miracle!"
"Color reproduction is a bloody miracle!"
AcknowledgmentsAcknowledgments
Peter BurnsDi-Yuan TzengGus BraunFrancisco Imai
Peter BurnsDi-Yuan TzengGus BraunFrancisco Imai