color reproduction and beyond...color reproduction and beyond roger david hersch ecole polytechnique...

Color Reproduction and Beyond

Roger David HerschEcole Polytechnique Fédérale de Lausanne (EPFL)

[email protected]

November 2013

1

Color reproduction: a solved problem ?

Nice color images are available everywhere, printable by anybody …

Is there foom for further research ?

Let us examine cases of color reproduction beyond classical cmyk

22

1. Case studies

A. Hiding patterns with metallic inks

B. Fluorescent color images visible only under UV light

C. Wide gamut printing with daylight fluorescent inks

D. Hiding watermarks with daylight fluo inks

E. Specular prints with classical inks on a metallic substrate

34At a non-specular angle At a specular angle

Ghost pattern

MetallicInk + classicalinks

A. Hiding patterns with classical and metallic inks

4

5

Metallic inks: reflection varies according to incident and viewing angles

Lightsource

Metallic ink

Paper

non-specularreflection

specularreflection

a a

How do you hide the patterns under non-specular reflection ?

MetallicArt

5 6

Principle

Prediction of reflectance spectrafor transparent inks

Prediction ofreflectance spectrafor superposed metallicand transparent inks

By fitting reflection spectracmy scsmsys

Rcmy R

cmy scsmsys

scsmsys

MetallicArt

[Hersch et al. 2003]6

7

Ghost patternappearingand disappearing

MetallicArt

7

B. Invisible fluorescent images

8

How does it work ?

Invisible fluorescent images

9 1010

Printing with invisible fluorescent ink halftonesYellow

fluorescent ink

Redfluorescent ink

Bluefluorescent ink

Paperblack


10

Emission spectraInvisible fluorescent images

11

Fluorescent inks & new colorantsDecrease concentration by reducing the printed dot size


12

1313

Emissive spectral prediction model

Halftoning: - colorants are placed side by side- paper black between colorant dots

=> Halftone emission spectra F() are sums of colorant emission spectra Fi() weighted by their effective surface coverages ui.

1( ) ( )

mi i

iF u F


[Hersch, Donzé, Chosson 2007] 13

Effective surface coverages

arg min ( ) ( )T

eff meas measu

u F u F F u F

ueff

0.5


14

Create the gamut and perform gamut mapping

Fluorescent ink gamut

Display gamut


15

a) Lightness mapping

b) Chroma mapping

Original

Photounder UV light

ResultsInvisible fluorescent images

16

C. Wide gamut prints with daylight fluo inks (mf , yf)

Classical Daylight fluorescent17

Reflection spectra of the fluorescent inks

To enlarge the gamut, how shall classical c,m,y,k inks and daylight fluo mf , yf inks be combined ?

Daylight fluo inks – wide gamut

[Rossier & Hersch 2012] 18

Gamut covered with daylight fluo inksFluorescent gamut Gf is union of 4 sub-gamuts: {cmfyf}, {cmfy}, {cmyf}, {cmyk}

1919

Gf and GsRGB gamuts

Parts of fluorescent gamut are larger than the sRGB gamut.How and when shall we perform gamut expansion ?

20


Multiple focii approach for gamut reduction and expansion

Avoid reducing too much chroma in dark tones21

Classical cmyk inks Classical and fluorescent inks cmyk + yfluo+ mfluo

Gauguin

R. Rossier, R.D. Hersch, Gamut expanded halftone prints, Proc. Color Imaging Conf, San Francisco, Nov. 2012

2222

D. Hiding watermarks with daylight fluorescent inks

Under daylight(photograph)

Under UV or blue light(photograph)

How do you hide patterns under daylight ?

Daylight fluo watermarks

23

c mf yf c m y

Prediction of reflectance spectrawith fluorescent inks

Prediction ofreflectance spectrawith classical inks

RcmyRcmyyf

daylightdaylight

Hide pattern by minimizing |Fit ink amounts c mf yf

Rcmf yf - Rcmy |


R. Rossier, R.D. Hersch, Hiding patterns with daylight fluorescent inks, Proc. IS&T/SID's 19th Color Imaging Conference, 2011, pp. 223-228.

E. Classical inks on metal

26

Large difference between specular and non-specularobservation:- Specular observation: high lightness, high chroma- Non specular observation: dark, low chroma

How do you compute the amount of each of the classical inks to create the target color halftonesobservable under specular reflection ?

Solution:

Establish a spectral prediction model for viewing colorsprinted on metal under specular reflection (25o:25o). Createthe full color reproduction workflow.

Specular prints with classical inks on metal

[Pjanic and Hersch 2013] 27

Map the display gamut to the printer

gamut

Establish the printer gamut

Characterize the printer (ink surface coverages printed CIELAB color)

2. Color reproduction workflow

Create ICC printer profile:display CIELAB color ink surface coverages

ICC profile

displaygamut

printergamut

Preparation:

28

Map input colors to ink surface coverages

Color reproduction workflow: image generation

Halftoning &printing (ink-jet, offset,..)

Color separations

Input imagecolors

Gamut mapping tabledisplay colors printable colors

Ink separation table: printable colors ink surface coverages

Possiblyembedded intoICC profile

& detect print variations (e.g. ink flow, dot gain, etc..)

[J.M. Lammens, Applications of Color Science and Technology in Digital Printer Research & Development, Proc. IS&T CGIV Conf, 2008, 1-6] 29

Input imagecolors

Color to ink surface coverage mapping

Halftoning &printing (ink-jet, offset,..)

Automatic detection of variations in printing conditions: dot gain and/or ink thickness changes

Spectral prediction modelred, green, blue, infrared

sensors

Deduce deviations in dot gain and/or ink thickness

R. D. Hersch et al., Deducing ink thickness variations by a spectral prediction model, Color Research and Application, Vol. 34, No. 6, Dec. 2009, pp 432-442 30

Alarm, recalibration

Prediction of sensor responses

3. Spectral prediction models

Goals in respect to printing:

- Predict the reflectance (->color) of a halftone patch as a function of ink surface coverages

- Fit surface coverages of cyan, magenta, yellow and black to match a desired spectrum/color ( characterization tables)

- Fit surface coverages of custom inks in order to match a desired color with a given optimization objective (maximize gamut, minimize metamerism, minimize usage of inks, minimize dot contrast..)

- Keep printer parameters constant (dot gain, ink thickness, ..)

31

Interaction between light, inks and paper

Coating & ink

Paper bulk

Air

(1) SpecularReflection (Fresnel)

(4) Internalreflections(Fresnel)

(3) Lateral scattering of light, reflection by paper bulk

(2) Transmission oflight through the ink layer

yields optical dot gain

32

The ink spreading phenomenon

Paper bulk

Ink dotTheory (nominal coverage)

Paper bulk

Ink dotPractice (effective coverage)

Solid ink

Solid ink

Ink spreading depends on the superposed paper and ink layers

Spectral prediction models

33

Classical sprectral reflection prediction models

Fundamental assumption: Colorants (=Neugebauer primaries) are formed by the paper white, the c,m,y inks and their superpositions (r,g,b,k) => 8 colorants

[Neugebauer 1937]

34

Colorant coverages deduced from inkcoverages (Demichel equations)

black (c & m & y)

cyanmagenta

green (cyan & yellow)blue (cyan & Magenta)

red (magenta & yellow)yellow

)1()1()1(

)1(

)1()1(

)1()1()1()1(

)1()1(

ymcwaymcka

ymcba

yycgaymcra

ymcyaymcma

ymcca

c, m, y : coverages of cyan, magenta & yellow ink dots

coverage of white

Spectral prediction models

35Equations valid for independently laid out c,m,y ink halftones

)()( i

ii RaRSpectral Neugebauer model

Ri : reflection spectrum of colorant i

ai : surface coverage of colorant i

does not account for lateral propagation of light

Sum of colorants weighted by their surface coverages

36

75 lpi100 lpi

150 lpi

E94

n

i

nii RaR

1

)()( a: surface coverage of ink

0.5

1

1.5

2

1.2 1.4 1.6 1.8 2 2.2 2.4 n

Prediction accuracy as a function of n :

Empirical approach[Viggiano 1990]

Yule-Nielsen modified Neugebauer modelNon-linear relationship between colorant reflection spectra and predicted reflection spectrum

Offset printing

n is fitted according to a set of measured patch reflectances.

37

4. Ink spreading models

0 0.25 0.5 0.75 1 -0.02

0

0.05

0.1

0.15

0.2

y

y/c

y/m

y/cm

y: yellow on paper

y/c: yellow on solidcyan

y/m: yellow on solidmagenta

y/cm: yellow on solidcyan and magenta

Physical dot gain : offset 150 lpi

nominal coverage

DotGain = effectiveCoverage – nominalCoverage

38

Ink spreading in every superposition conditions

Separate mappings of nominal to effective dot coverages for (a) inks printed alone on paper (b) inks superposed with one solid ink (ink spreading)(c) inks superposed with two solid inks (ink spreading) 39

Calibration of ink spreading curves (3 inks) Reflectances of primaries (solid colorants formed by inks & ink superpositions)

Reflectances of single ink halftones in all 12 superposition conditions

50% c,m,y on paper

50% cyan on m, y, m&y (red)

50% magenta on c, y, c&y (green)

50% yellow on c, m, c&m (blue)

c m y r g b k w

40

Resulting 12 ink spreading curves (3 inks)

41

75%

For 3 inks 12 superposition conditions With 50% patches only: 8 + 12 = 20 patchesWith 25%, 50%, 75% surface coverages:

8 solid + 12 x 3 halftones = 44 patchesFor 4 cmyk inks: 20 superposition conditionscmyk with halftone black: 16 solid + 20 halftones = 36 patches[Bugnon, Brichon & Hersch 2008: eliminate halftones on solid black]

In general for 4 inks: 32 superposition conditionscmyk: 16 solid + 32 halftones = 48 patches

Number of calibration patches

42

Weighting the surface coverage mapping functions

/

/

/

c w

mc m

c y y

c my my

wc' f (c)

f (c) wwf (c)

wf (c)

= +

+

+

ww : prop. of cyan over white

wm : prop. of cyan over magenta

wy : prop. of cyan over yellow

wmy : prop. of cyan over red(magenta + yellow)

[Hersch & Crété 2005]

43

Weighting the surface coverage mapping functions

( ) ( ) ( )( ) ( )( ) ( )( )

' 1 ' 1 - '

' 1 '/

1 ' '/

' '/

c f c m y c

f c m y c mf c m y c yf c m y c my

= - +

- +

- ⋅ +

prop. of white

prop. of magentaprop. of yellow

prop. of red

Effective coverage of cyan:

44

R. D. Hersch, F. Crété, Improving the Yule-Nielsen modified spectral Neugebauermodel by dot surface coverages depending on the ink superposition conditions, SPIEVol. 5667, 434-445 (2005)

fc(c)

y

fy(y) fm/y(m) fy/c(y) fy/m(y)

c

fy/cm(y)

c' m' y'

aw ' ac ' am ' ay ' ag 'ar ' ak '

Predicted spectrum

m

fm(m)

YNSN spectral prediction model

linear combination according ratio of colorant coverages

Demichel equations

Nominal ink surface coverages

Colorant reflectancespaper reflectance

ab '

Effective ink surface coverages

Prediction framework with ink spreading in all superposition conditions

fc/m(c) fc/y(c) fc/my(c)fm/y(m) fm/cy(m)

Effective colorant coverages

45Prediction accuracy: E94 : 0.9 to 1.5

5. Ink spreading enhanced Cellular Yule-Nielsen(a) Subdivide the nominal ink surface coverage space

for 3 inks 8 subcubes; for 4 inks -> 16 subcubes

46

Cellular Yule-Nielsen: normalization within cells (here 3 inks)

(b) Normalize the nominal surface coverages within each subcube

' ' 'l l l

h l h l h l

c c m m y yc m yc c m m y y

(c) Predicted reflectance by tri-linear interpolation of subcube vertex reflectances

1/ 1/, , , ,

1/ 1/, , , ,

1/ 1/, , , ,

,

( )

(1 ')(1 ')(1 ') ( ) '(1 ')(1 ') ( )

(1 ') '(1 ') ( ) (1 ')(1 ') ' ( )

(1 ') ' ' ( ) '(1 ') ' ( )

' '(1 ')

pred

n ncl ml yl ch ml yl

n ncl mh yl cl ml yh

n ncl mh yh ch ml yh

ch m

R

c m y R c m y R

c m y R c m y R

c m y R c m y R

c m y R

1/ 1/, , ,( ) ' ' ' ( )

nn nh yl ch mh yhc m y R

47

Cellular Yule-Nielsen: accounting for ink speading

One ink spreading curve per ink and per subcube:

Fit the 0.5 normalized nominal surface coverage and pass a parabola through it.

Fit coveragesRcenter()

Rcl,ml,yl() Rcl,mh,yl()

Rch,mh,yl()

Rch,mh,yh()Rch,ml,yh()

Rcl,ml,yh() Rcl,mh,yh()

48

R. Rossier, T. Bugnon, R.D. Hersch, Introducing ink spreading within thecellular Yule-Nielsen modified Neugebauer model, Proc. IS&T 18th ColorImaging Conference, 295-300 (2010).

Cellular Yule-Nielsen: accounting for ink speading

Fit simultaneously all 3 effective ink dot surface coverages at the center of cell, normalized nominal surface coverages of (0.5,0.5,0.5)

2

ˆ ˆ ˆ', ', '

ˆ ˆ ˆ', ', '

ˆ ˆ ˆ( , ', ', ') ( , ', ', ')arg min j j j

j j jpredmeas

j k j j j k j j jjc m y k

c m y

R c m y R c m y

Number of calibration patches for 3 inks: 27 primaries + 8 subcube centers = 35 patches

49Prediction accuracy: E94 : 0.5 to 1.2

125 test samples Mean Max Quant 95% Mean Max Quant 95%

n=IS-YNSN | CYNSN 50lpi, n = 2 | 2 75lpi, n = 3 | 2IS-YNSN, ink spreading in all superposition conditions 0.9 2.6 1.75 0.9 2.1 1.9

Cellular YNSN, single inkspreading curve per ink 0.6 2.2 1.5 0.6 2.1 1.4

IS-YNSN, ink spreading in all superposition conditions 0.9 2.1 1.6 1.0 2.3 1.77


94ΔE 94ΔE 94ΔE 94ΔE

Prediction accuracy: cmy inkjet Epson P50high densities (1.2 to 1.4)

100lpi, n = 5 | 4 125lpi, n = 14 | 3

94ΔE 94ΔE

Epson P50, solid inks, high densities cmy = {1.4, 1.2, 1.46 }IS-YNSN: 44 calibration patches, CYNSN: 35 calibration patches 50

6. What’s next ?

3D printingCreation of a layer having predetermined surface facet shapes

51

T. Weyrich, T., P. Peers, W. Matusik, S. Rusinkiewicz, Fabricating microgeometry for custom surface reflectance. ACM Transactions on Graphics, (28) 3 (2009).

Possibility of printing a 2.5D relief

52

Shadow image with red, green & blue light sources

Shadowimagewith single light source

M. Alexa, W. Matusik, Reliefs as images.ACM-Trans. on Graphics, Vol. 29, No. 4,60:1–60:7 (2010)

Printing 2.5D brush strokes

C. Parraman, The development of vector based 2.5D print methods for apainting machine, Conf Color Imaging XVIII: Displaying, Processing,Hardcopy, and Applications, SPIE Vol. 8652, paper 86520R, 1-8 (2013)

54

base bands

revealer: transparent line grating

Synthesis of a 1D moiré with transparent lines or lenticular lenses

R.D. Hersch, S. Chosson, Band Moiré Images, ACM Trans. on Graphics (Proc. SIGGRAPH), Vol. 23, No.3, 239-248 (2004)

53

Revealing layer made of lenticular lenses

54

Circular Moiré image

55

6. ConclusionsWe extended classical CMYK prints :- silver inks- invisible fluorescent inks- daylight fluorescent inks- specular reflecting substrates (prints on metal)These prints still require a color reproduction workflow with- a spectral prediction model- gamut mapping- halftoning

Future: 2.5D printing enable- creating relief structures- composition with diffuse and transparent substrates- optical effects with micro-lenses

Challenge: color halftoning in 3D 56

ThanksFrom EPFL:I. Amidror, P. Amrhyn, V. Babaei, M. Brichon, T. Bugnon, S. Chosson, F. Collaud, F. Crété, P. Donzé, P. Emmel, P. Fehr, S. Jain, H. Janser, N. Garg, M. Hebert, A.K. Singla, S. Mourad, V. Ostromoukhov, P. Pjanic, R. Rossier, N. Rudaz, R. Seri, T. Walger.

From industry: M. Riepenhoff, H. Janser, Wifag AG, BernP. Maturo, H. Ravez, H. Bouchrara, Perfect AG, EtoyM. Meyer, S. Gasser, Genoud SA. Le MontS. Lizzola, SSP SA, Lausanne& others from security companies

ReferencesM. Alexa, W. Matusik, Reliefs as images. ACM-Trans. on Graphics, Vol. 29, No. 4,

60:1–60:7 (2010)V. Babaei, R. Rossier, R.D. Hersch, Reducing the number of calibration patterns for

the two-by-two dot centering model, Conf. Color Imaging XVII: Displaying,Processing, Hardcopy, and Applications, SPIE Vol. 8292, 2012, paper 829208,pp. 1-9

V. Babaei, R.D. Hersch, Juxtaposed Color Halftoning Relying on Discrete Lines,IEEE Transaction on Image Processing, Vol. 22, No. 2, 2013, 679-686

R. Balasubramanian, Optimization of the spectral Neugebauer model for printercharacterization, Journal of Electronic Imaging, Vol. 8, No. 2, 156-166 (1999)

F. Bernardini, J. Mittleman, H. Rushmeier, C. Silva, and G. Taubin, The Ball-Pivoting Algorithm for Surface Reconstruction, IEEE Trans. Visualization andComputer Graphics, vol. 5, no. 4, pp. 349-359, Oct.-Dec. 1999

G. V. J Cadarso, S. Chosson, K. Sidler, R. D. Hersch and J. Brugger, Light: Science& Applications (2013) 2, e86; doi:10.1038/lsa.2013.42, pp. (1-5) July 2013

T.J. Cholewo and Love S., “Gamut boundary determination using alpha-shapes”, Proc. IS&T/SID's 7th Color Imaging Conference, 1999.

F.R. Clapper, J.A.C Yule, The effect of multiple internal reflections on the densitiesof halftone prints on paper, J. of the Optical Society of America, Vol. 43, 600-603(1953)

57

Th. Bugnon, M. Brichon, R.D. Hersch, Model-based deduction of cmyk surfacecoverages from visible and infrared spectral measurements of halftone prints,Conf. Color Imaging XII, SPIE-IS&T Electronic Imaging, SPIE Vol. 6493, pp649310-1 to 649310-10 (2007)

Th. Bugnon, M. Brichon, R.D. Hersch, Simplified Ink Spreading Equations for CMYKHalftone Prints, Conf. Color Imaging XIII, SPIE-IS&T Electronic Imaging, SPIEVol. 6807, pp 680717-1 to 680717-12 (2008)

R .D. Hersch, P. Emmel, F. Crété, F. Collaud, Spectral reflection and dot surfaceprediction models for color halftone prints, J. of Electronic Imaging, Vol. 14, No.3, August 2005, 33001-12

R. D. Hersch, F. Crété, Improving the Yule-Nielsen modified spectral Neugebauermodel by dot surface coverages depending on the ink superposition conditions,Electronic Imaging Symp., Conf. Imaging X: Processing, Hardcopy andApplications, SPIE Vol. 5667, 434-445 (2005)

R. D. Hersch, F. Collaud, P. Emmel, Reproducing color images with embedded metallic patterns, Proc. SIGGRAPH ACM Trans. on Graphics, Vol. 22, No.3, 2003, pp. 427-436

R.D. Hersch, S. Chosson, Band Moiré Images, ACM Trans. on Graphics (Proc. SIGGRAPH), Vol. 23, No.3, 239-248 (2004)

R.D. Hersch, P. Donzé, S. Chosson, Color images visible under UV light, Proc. Siggraph 2007, ACM Trans. Graph. vol. 26, No. 3, Article 75, 9 pages, 2007

R. D. Hersch, M. Brichon, T. Bugnon, P. Amrhyn, F. Crété, S. Mourad, H. Janser, Y.Jiang, M. Riepenhoff, Deducing ink thickness variations by a spectral predictionmodel, Color Research and Application, Vol. 34, No. 6, Dec. 2009, pp 432-44258

R.D. Hersch, Spectral prediction model for color prints on paper with fluorescentadditives, Applied Optics, Vol. 47, No. 36, 2008, pp. 6710-6722

J.M. Lammens, Applications of Color Science and Technology in Digital PrinterResearch & Development, Proc. IS&T CGIV Conf, 2008, 1-6

J. Morovic, M. R. Luo, The fundamentals of gamut mapping: A survey, J. Imag. Sci.Technol., vol. 45, no. 3, 2001, pp. 283–290.

H.E.J. Neugebauer, Die theoretischen Grundlagen des Mehrfarbendrucks. Zeitschriftfuer wissenschaftliche Photographie, Vol. 36, 36-73, (1937), reprinted inNeugebauer Seminar on Color Reproduction, SPIE Vol-1184,194-202 (1989)

C. Parraman, The development of vector based 2.5D print methods for a paintingmachine, Conf Color Imaging XVIII: Displaying, Processing, Hardcopy, andApplications, SPIE Vol. 8652, paper 86520R, 1-8 (2013)

P. Pjanic and R. D. Hersch, Specular Color Imaging on a Metallic Substrate, 21stColor and Imaging Conference 2013, in press

J. P. Reveillès, “Combinatorial pieces in digital lines and planes,” SPIE VisionGeometry IV, 2573 1995, 23–34.

R. Rossier, T. Bugnon, R.D. Hersch, Introducing ink spreading within the cellularYule-Nielsen modified Neugebauer model, Proc. IS&T 18th Color ImagingConference, 295-300 (2010).

R. Rossier, R.D. Hersch, Hiding patterns with daylight fluorescent inks, Proc.IS&T/SID's 19th Color Imaging Conference, 2011, pp. 223-228.

R. Rossier, R.D. Hersch, Gamut expanded halftone prints, 20th Color and ImagingConference Final Program and Proceedings CIC Conf, 2012, pp. 315-322 59 60

J.A.S Viggiano, Modeling the Color of Multi-Colored Halftones, Proc. TAGA, 44-62 (1990)

S.G. Wang,Two-by-Two Centering Printer Model with Yule-Nielsen Equation, IS&T International Conference on Digital Printing Technologies, 1998, 302-305.T. Weyrich, T., P. Peers, W. Matusik, S. Rusinkiewicz, Fabricating microgeometry for custom surface reflectance. ACM Transactions on Graphics, (28) 3 (2009).

J.A.C. Yule, W.J. Nielsen, The penetration of light into paper and its effect on halftone reproductions, Proc. TAGA, Vol. 3, 65-76 (1951),






This image cannot currently be displayed.This image cannot currently be displayed. This image cannot currently be displayed.

Prediction accuracy: cmy inkjet Epson P50high densities (1.2 to 1.4)

100lpi, n = 5 | 4 125lpi, n = 14 | 3

This image cannot currently be displayed.


Epson P50, solid inks, high densities cmy = {1.4, 1.2, 1.46 }IS-YNSN: 44 calibration patches, CYNSN: 35 calibration patches 61





Cellular YNSN, single inkspreading curve per ink 0.4 2 1.4 0.4 1.9 1.2

This image cannot currently be displayed.This image cannot currently be displayed.

This image cannot currently be displayed.This image cannot currently be displayed.

Moderate densities (~1)

100lpi, n = 4 | 6 125lpi, n = 5 | 9


Epson P50, solid inks, moderate densities cmy = {0.95, 0.77, 1.04} 62

Ink Spreading Model for 4 inks

32 ink spreading curves

Yellow Blacky y/k k k/yy/c y/ck k/c k/cyy/m y/mk k/m k/myy/cm y/cmk k/cm k/cmy

Cyan Magentac c/k m m/kc/m c/mk m/c m/ckc/y c/yk m/y m/ykc/my c/myk m/cy m/cyk

Assumption: Any ink on black yields black Remove all ink

spreading curvesover solid black

Ink spreading models

63

Simplified ink spreading equations for CMYK

Remove the ink spreading curves on solid black

/ /

/ /

/ /

//

//

/

' 1 ' 1 ' ' 1 ' 1 '' 1 ' ' 1 '

1 ' ' 1 ' '' ' ' '

' 1 ' 1 ' ' 1 ' 1 ' 1 '' 1 ' 1 '' 1 '

1 ' ' 1 '1 ' ''' '

c m

c m m c

c y m y

c my m cy

y k

k cy c

k my m

y cm

c m y f c m c y f mm y f c c y f m

m y f c c y f mm y f c c y f m

y c m f y k c m y f kc m y f kc m f y

c m y f kc m f ycc m f y

/

/

/

/

/

' 1 '1 ' 1 ' '

' 1 ' '1 ' ' '

' ' '

k cm

k y

k cy

k my

k cmy

m y f kc m y f k

c m y f kc m y f k

c m y f k

[Bugnon, Brichon, Hersch 2008]

Prediction accuracy: E94 : 0.9 to 1.5

color reproduction and beyond...color reproduction and beyond roger david hersch ecole polytechnique...

Documents