Color - Science, Management, and Quality 2008 1 3folk.uio.no/inf5300/2008/Color-JYH-2008.pdf ·...
Transcript of Color - Science, Management, and Quality 2008 1 3folk.uio.no/inf5300/2008/Color-JYH-2008.pdf ·...
1
Color Science, Color Management, and Color Image QualityProfessor Jon Yngve Hardeberg
The Norwegian Color Research Laboratory and Hardeberg Image andGjøvik University College, Gjøvik, Norway Color Technology
[email protected], http://www.colorlab.no, http://color.hardeberg.com
3
IntroductionYou process images
Maybe capturing them Maybe trying to improve their visual appearanceMaybe you want to compress them without sacrificing qualityMaybe you try to recognize objects, something humans do easilyMaybe you want to visualize complex image data using color?
4
IntroductionYou need to
Understand color perceptionMake sure What You See Is What Others SeeBe able to judge image quality
5
OutlineIntroductionColor ScienceColor ManagementColor Image Quality
6
What is Color?
Well, my favorite color is red!
(t-t-t togduan?)
7
What is Color?
Well, my favorite color is red!
“Color consists of the characteristics of lightother than spatial and temporal inhomogeneities;
light being that aspect of radiant energy of which a human observer is aware through the
visual sensations which arise from the stimulation of the retina of the eye.”
[OSA 1940]
2
8
What is Color?
Color is a sensationInteraction between the physical world and our sensesPsychophysics
“Color consists of the characteristics of lightother than spatial and temporal inhomogeneities;
light being that aspect of radiant energy of which a human observer is aware through the
visual sensations which arise from the stimulation of the retina of the eye.”
[OSA 1940]
9When a tree falls in the forest, and no one is around to hear it, is there a sound?
Is the tree green if no one’s there?
"I want you to realize that there exists no color in the natural world, and no sound -nothing of this kind; no textures, no patterns, no beauty, no scent." (Sir John Eccles)
10
Color: Light, surface, eye
Wavelength, λ
Rad
ianc
e, l(
λ)
Light’s Spectral Distribution, l(λ)
Wavelength, λ
Ref
lect
ance
, r(λ
)
Surface’s Spectral Reflectance, r(λ)Eye’s Spectral Sensitivities si(λ), i =1,2,3
Wavelength, λ
Sens
itivi
ty, s
i(λ) S M
L
11
Stare at this flag for 30 seconds...
12 13
What did you see?
#1: A white screen only
#2: The Norwegian flag
#3: The Danish flag
#4: First #3 then #2
Complementary after-images
ConclusionsThe eye is quite different from a CCD sensor Our linear model is limited (but very useful)
3
14
Retinal cone distribution vs. CMOS sensor15
ColorimetryInternational LightingCommission – CIE
1931 Standard Colorimetric ObserverDetermined based on colormatching experiments
Provides a standardizedway of quantifying color
CIEXYZFoundation for consistentcommunication of color
16
Color Spaces3 different photoreceptors, L, M, S
3-output linear modelColor can be specified by three numbersColor space - analogy to geometrical 3D space
C2
C3
C1
C
17
RGB space is most commonNative to cameras, scanners, displaysBased on additive color mixing
CMY(K) for printingSubtractive color mixing
Luminance-chrominancespaces
Important to separate, for instance for ”Chroma Subsampling”HSV, HSL, YIQ, YUV, YCbCR, Lab, Luv, …
Color spaces for images and video
18
YCbCr Color SpaceA luminance-chrominance color space: YCbCr
Standardized for digital video by Recommendation ITU-R BT.601.4 Used in JPEG image compression and MPEG video compression.Closely related to the YUV transform.
Luminance Y’ (a.k.a. “Luma”)Y’ = 0.299R’ + 0.587 G’ + 0.114B’Calculated from gamma-corrected signals R’, G’, B’Note difference from colorimetric luminance Y
Chrominance Cb and CrCb = ((B’−Y’) / 1.772) + 0.5Cr = ((R’−Y’) / 1,402) + 0.5
19
Color Spaces3 different photoreceptors, L, M, S
3-output linear modelColor can be specified by three numbersColor space - analogy to geometrical 3D space
C2
C3
C1
C
Device-independent color spacesLMS, XYZ, CIELUV, CIELAB, sRGB …
Device-dependent color spacesMonitor RGB, Printer CMYK, Scanner RGB ...
4
20
Device independenceDevice-independent color spaces
Defined according to colorimetryLMS, CIEXYZ, CIELUV, CIELAB, sRGB …
Device-dependent color spacesMonitor RGB, Printer CMYK, Scanner RGB ...
21
CIELAB color space (a.k.a. L*a*b*)Device independentPseudo-uniformly related to human perception:
Defined according to a standard observerInternational Standard (CIE) Used in Color Fax and CMS
Lightness L*L* = 100 = white L* = 0 = black
Chromaticity a* and b*a* = green (-) to red (+)b* = blue (-) to yellow (+)
22
sRGB Color SpaceStandardized RGBDevice independentDefinition based on a typical PC monitorAdvantages:
Ease of useInternational Standard (IEC)Endorsed by major business players (HP, MSFT)A common color language - Scan to sRGB - Print from sRGB - - -More and more consumer imaging devices “speak” sRGB
Disadvantages:Yes
http://www.srgb.com
From
http:
//www
.srgb
.com
23
Color DifferenceEuclidean distance in CIELAB space
ΔE*ab “Delta E”A good measure of perceptual color difference Much used as color quality metric:
Original Reproduction
Average ΔE*ab = ?Max ΔE*ab = ?
24
Perceived color depends on...Spectral distribution of lightSpectral reflectance of surfaceSpectral sensitivity of eye
… but also on …
Size of areaViewing directionViewer adaptationSurrounding colorLocal contrast …
Brain
25
5
26
Assimilation27
Simultaneous contrast
28 29
30 31
Local vs. global color changes
2
2
2 Adding a cyan filter over the cushion
3
3
3 Adding a cyan filter over the whole image
Illustrates adaptation to white point -color constancyWhite balance for camerasColor appearance modeling - CAM
1
1
1 Original image
Images courtesy of Dr. R.G.W Hunt, Univ. of Derby, UK
6
32
OutlineIntroductionColor ScienceColor ManagementColor Image Quality
33
The first law of color managementDifferent imaging devices never produce equal color!
34
Scanners ‘speak’ color differently
File B
File AScanner A
Scanner B
OriginalImage
≠
35
Printers ‘speak’ color differently
Printer A
Printer B
Digital file
≠
36
We speak color differently
I wonder if this skirt matches mybeautiful “Barbie Doll Pink” blouse?
Let’s see - they write it’s “Hot Pink,”and in the image it says
(R,G,B) = (245, 174, 203) ??!?
37
Complex imaging systems
7
38
Principle of Color ManagementImage interchange through Device-Independent Color Space
Esperantoof colors
Imaging devices characterized by Profiles
Dictionaries
ScannerProfile
Device-Independent Color Space
PrinterProfile
ImageProfile
MonitorProfile
CameraProfile
39
Principle of Color Management
ScannerProfile
PrinterProfile
Color conversion between devices by “coupling” two profiles - source and destination
40
LUT-based colorspace conversion
G
B
R
One look-up table (3DLUT) per conversion
Pre-calculated conversion for a subset of all possible input colors
Interpolation for in-between valuesContent of LUTs depends on the actual devices, and is determined by device profiling
A.k.a colorimetric characterization
41
Colorspace conversions in an MFP• Local color copy
• Scanner RGB Printer CMYK
• Scan to PC
• Scanner RGB sRGB
• Print from PC
• sRGB Printer CMYK
• ITU Color Fax
• Scanner RGB CIELAB
• CIELAB Printer CMYK
• Internet Color Fax
• Scanner RGB YCbCr
• YCbCr Printer CMYK
Scan Device
ColorspaceConversions
CIELAB
PCApplication
sRGB
CMYK
PSTNColor Fax
YCbCr
Internet Color Fax
Print Device
ScannerRGB
42
International Color Consortium - ICCInternational standard for Color Management
Architecture of Color Management SystemsICC Profile file formatNow also ISO standardSee www.color.org
43
Input Device Profiling (scanner)
Previously known or measured
colorimetric dataScanner Profiling
AlgorithmColor Target
Mean values extraction
ScannedImage
File
ICC Profile
Device Independent
ColorImage
Scanner PC
Device-dependent RGB data
Device-independent color data (CIELAB)
8
44
Output Device Profiling (printer)
PrinterProfiling
CMYK or RGBprinter target
digital file
Device-dependent CMYK or RGB data
Device-independent color data (CIELAB)
Image File from Fax or
Scan
PrintingEngine
PrinterController
ICC Profile
45
Approaches to device profilingA.k.a colorimetric device characterizationTypical algorithms based on
Physical modelsGOG model for CRT displaysNeugebauer model for halftone print++
Data-driven models3DLUT InterpolationPolynomial regressionNeural networksThin-plate splines++
46
Some active research areasImproved algorithms for deviceprofiling / colorimetriccharacterization
LCD and DLP projection displaysHue-preserving cameracharacterization model
Compensation for varyingviewing conditions
Color appearance modelingColor gamut mappingSpectral reproductionWorkflow issues
Ease of useGet people to do the right thingsNew areas such as digital film production
C.F. Andersen and J.Y. Hardeberg, Colorimetric characterization of digital cameras preserving hue planes, 13th ColorImaging Conference, pp. 141-146, 2005
L. Seime and J.Y. Hardeberg, Colorimetric Characterisation of LCD and DLP Projection Displays, Journal of the Society of Information Display, 11(2): 349-358, 2003
J.Y. Hardeberg, Colorimetric ScannerCharacterization, Acta Graphica, 155, 2005
J.Y. Hardeberg, I. Farup, Ø. Kolås, and G. Stjernvang, Color management in digital video: Color Correction in the Editing Phase, Proc. IARIGAI Conference, pp. 166-179, Lucerne, Switzerland, September 2002
E.B. Mikalsen; J.Y. Hardeberg & J.-B. Thomas. Verification and extension of a camera-based end-user calibration method for projection displays. CGIV, 2008
A. Alsam; J. Gerhardt & J.Y. Hardeberg. Inversion of the Spectral NeugebauerPrinter model. AIC Colour 05, 44-62, 2005
47
Color Gamut MappingImaging devices have only a limited number of reproducible colors: Color GamutWhat to do if your image contains “out-of-gamut” colors?
Can’t print anything “more yellow” than the yellow ink!
Need for Gamut Mapping“Do the best you can with the colors you have”Huge unsolved problem in color imaging R&DClipping out of gamut colors is not good enough
48
Gamut Mapping and VisualizationCurrent research topics
Image-dependent and/or interactive new gamut mapping algorithmsImportance of gamut boundary descriptors:
Segment-maxima, convex hull, alpha-shapes, regular structure, …
Spatial gamut mapping With University of Milano
Popular freeware tool developed by us
ICC3D http://www.colorlab.no/icc3d/ I. Farup, J. Y. Hardeberg, A. M. Bakke, S.
Kopperud, and A.Rindal, Visualization and Interactive Manipulation of Color Gamuts, Proc. 10th Color Imaging Conference, pages 250-255, Scottsdale, Arizona, November 2002
Ivar Farup, Jon Y. Hardeberg, and MortenAmsrud, Enhancing the SGCK Colour Gamut Mapping Algorithm, Proc. CGIV 2004, pp. 520-524, Aachen, Germany, April 2004
Ivar Farup and Jon Y. Hardeberg, Interactive Color Gamut Mapping, Proceedings of the 11th International Printing and Graphic Arts Conference, Bordeaux, France, October 2002
Ivar Farup; Carlo Gatta & Alessandro Rizzi. A Multiscale Framework for Spatial Gamut Mapping. IEEE Transactions on Image Processing, 16(10):2423-2435, 2007
Øyvind Kolås & Ivar Farup. Efficient Hue-preserving and Edge-preserving Spatial Color Gamut Mapping. IS&T and SID's 15th Color Imaging Conference, 207-212, 2007.
Arne Magnus Bakke; Jon Yngve Hardeberg & Ivar Farup. Evaluation of Gamut BoundaryDescriptors. IS&T and SID's 14th ColorImaging Conference 50-55, 2006
49
OutlineIntroductionColor ScienceColor ManagementColor Image Quality
9
50
What is color image quality?51
What is color image quality?
52
What is color image quality?
53
What is color image quality?
What is color?What is an image?What is quality?
Dictionary: “Degree of excellence”ISO9000: “The features of a product (or service) which are required by a customer”
54
Quality is what the customer wants!Quality is what the customer wants!Quality is what the customer wants!Quality is what the customer wants!Quality is what the customer wants!Quality is what the customer wants!Quality is what the customer wants!Quality is what the customer wants!
55
Customer-driven quality definitionNorwegian governmental purchase of digital print equipmentE.g. letterhead quality of primordialimportance
P. Nussbaum and J.Y. Hardeberg, Print Quality Evaluation for Governmental Purchase Decisions, Proc. IARIGAI Conference, Copenhagen, September 2004
10
56
How to evaluate color image quality?57
How to evaluate color image quality?From the web site of a major French consumer electronics retailer:
Qualité d'image = taille du point x nuances de couleursImage quality = point size x color depth
NOT
58
How to evaluate color image quality?NOT 59
How to evaluate color image quality?NOT
60
How to evaluate color image quality?
Subjective evaluations”Expert opinions”Customer’s choices
Psychophysical experiments”Panel tests”Statistical analyses
MeasurementsColorimetric and planimetricPixel-by pixel comparisonsPerceptual models etc.
61
Psychophysical experimentsDifferent protocols
Category judgmentRank orderPair comparison
Time consumingThorough statisticalanalysis
But some questionableassumptions?
11
62
Measuring color image qualityBy color measurements
Match of specific colorsE.g. Company logo
Match monitor/printTypical consumer user scenario
Match original-reproductionColor copy, fax
Match to a reference print
Do not use numbers blindly!
Original
Reproduction
Mean ΔE*ab = ?Max ΔE*ab = ?
63
Preferred color reproductionGenerally it is not desired to reproduce the colorimetry of the original scene
(Gio
rgia
nni&
Mad
den,
199
7)
64
Colorimetric match ≠ visual matchBasic colorimetry is inadequate for cross-media color reproduction
65
Inadequacy of RGB color difference
ΔRGB=100 ΔRGB=100
ΔEab=22
CIEDE2000=4,8
ΔEab=63
CIEDE2000=32
66
Reproduction 1 and 2 have the same average ΔEab of 0.6CIE ΔEab colour difference equation is designed to quantify the colour difference for single pair of colour patchesDon’t ever use RMS or PSNR to evaluate yourimage compression algorithm again!
35Eab =Δ
Reproduction 1
Inadequacy of pixel-by-pixel difference
Original Reproduction 2
67
Perceptual image difference metricsExactly how big is the difference betweenthese two images?
12
68
Perceptual image difference metricsMetrics taking into account visual CSFs have beenproposed
s-CIELABiCAM
Many othermetrics
SSIM
69
Perceptual image difference metrics
Image difference maps
S-CIELAB iCAM CIELAB
ReproductionOriginal
70
Perceptual image difference metricsWe found no correlationbetween current image difference metrics and perceptually evaluatedimage differencesHigher-level cognitiveprocesses come into playCurrent active researcharea
E. Bando, J.Y. Hardeberg, D. Connah, Can Gamut Mapping Quality be predicted using Colour Image Difference Formulae?, SPIE Proc. 5666, 2005
71
Perceptual image difference metricsResearch project funded by theResearch Council of Norway
Ca 5.4 MNOK over 4 years 2007-2011
Related project funded by Océ PrintLogic Technologies (Paris, France)
Ca 1.4 MNOK over 4 years 2007-2011
New metrics development/evaluationImplement and evaluate existingmethodsNew metrics based on improvedcolour difference metricsNew metrics based on perceptualpredictors
Such as RetinexNew metrics based on saliency maps
Eye tracking experimentsNew metrics based on multilevelmethods
Using metrics to improvereproduction&representation
Standard optimization methodsHeuristic methodsVariational methods
ApplicationsImage compressionHigh dynamic range tone mappingColour gamut mappingHalftoning
72
Perspectives on image quality research”Most studies on image quality employ subjectiveassessment with only one goal – to avoid it in the future”
Yendrikhovskij
CIQmachine
CIQ=100!CIQmachine
73
OutlineIntroductionColor ScienceColor ManagementColor Image Quality
13
74 75
The Norwegian Color Research Laboratory
Research group at the Faculty of Computer Science and Media Technology at Gjøvik University College
Rooted in graphic arts industry needs but now serving the converging media technology industry
Great people3 permanent faculty3 associated faculty4 doctoral students and 1 postdoctoral researcher1 color management engineerBachelor and master students
Well equipped laboratory facilitesColor and spectral measurement, Color managementsoftware, viewing booths, image acquisition and reproduction devices, monochromator, etc. >200m2 lab space
Research in color science and image processing
Color Management and Device CharacterizationColor Image QualityColor Gamut Mapping and VisualizationMultispectral Color Image Acquisition and ReproductionFundamental Colorimetry and Psychophysics
Strong ties to the faculty’s media related bachelor and master programs
http://www.hig.no/mediahttp://www.master-erasmusmundus-color.eu/
Rich national and international network with industry and academiahttp://www.colorlab.no
76
Colorlab Research OverviewApplied research
Graphic arts industryColor managementPrint quality evaluationPSO color certification
Media technology industryDisplay technologyMovie productionPhoto managementTeleconferencing (?)
Computer visionRecycling sorting
Security applicationsPeople countingIntelligent videosurveillanceDocument and imageforensics
Fundamental researchMultispectral colorimaging
NFR/SHP 2003-20074.8 MNOK
Perceptual image difference metrics
NFR/SHP 2007-20115.51 MNOK
Color and perceptionColor memoryEye movements
In-between…Gamut mappingPrint and image qualityDevice modelingFace detection
77
CREATEColor Research for European AdvancedTechnology Employment
http://www.create.uwe.ac.uk/Series of conferences and research courses
Print-oriented courses in the UK, fall 2008Large final conference in Gjøvik in 2010
EU- funded through Marie Curie programFree participation (including travel) for selectedyoung researchers
Partners:Università degli Studi di Milano, Italy; Gjøvik University College, Norway; University of Leeds, UKUniversity of Ulster in Belfast, Northern Ireland, Universidad Autónoma de Madrid, Spain; Université de Reims Champagne-Ardenne, France; University of Pannonia, Hungary.
78
CIMET key factsMSc programme: Color in Informatics and Media TechnologyCollaborative effort supported by the EU
In Master Mundus programme: bringing together the best Europan forces to attract the best brains to EuropeUniversity of Saint-Etienne (France, coordinator), the University of Granada (Spain), the University of Joensuu(Finland), and Gjøvik University College (Norway)
Prestigeous for institutionsOnly three selected programmes in Norway in 2007
Prestigeous for studentsGenerous grants from EU
To cover tuition fees (EUR 10.060 per year)Highly competitive selection
Only 30 students per year
http://www.master-erasmusmundus-color.eu/
79
Questions?E-mail:
[email protected]@hardeberg.com
Web:http://www.hig.nohttp://color.hardeberg.comhttp://www.colorlab.no
Thanks to:Current and past Colorlab faculty and students
Upcoming events:Open Colorlab Workshops
4.6.08: Psychophysical experiments and statistics19.6.08: Face detection and recognition
Gjøvik Color Imaging Symposium 2009Immediately following SCIA09