Color Reproduction / Color Gamut Mapping: Two Sides of - Create!

John McCann, McCann Imaging, USA.

Color Reproduction / Color Gamut Mapping:Two Sides of the Same Coin

Tuesday, October 20, 2009

Outline

• Color Reproduction & Gamut Mapping

• 3-D color spaces [One into Another]

• Practice

• Reproduce Paintings

• Gamut Mapping

• Theory


But first,CREATE

Colour Research for European Advanced Technology

Employment

New Ways to use Print Technology : (On Paper)

School of Creative Arts, University of the West of England Bristol (UK)

14 - 17th October, 2008


Allesandro Rizzi

Dip. Tecnologie dell’Informazione,Università degli Studi di Milano,

Italy

30

Carinna Parraman

Centre for Fine Print Research, University of the West of England,

UK


LDR

HDR

Measure appearances of constant reflectancesin different illuminations


Colour Constancy

Reflectance

Illumination

Eye

Eye Radiance= (Illumination * Reflectance)


Colour Constancy

Reflectance

Illumination

Eye

Eye Radiance= (Illumination * Reflectance)

Discount Illumination

SynthesizeReflectancefrom Edges


1968


Reflectances

Long = 21Middle = 62

Short = 8

Long = 85Middle = 48Short = 67


Illuminations

Long = 85Middle = 48Short = 67

Long = 21Middle = 62

Short = 8


Mondrians Long = 53Middle = 55Short = 38


Mondrians Long = 53Middle = 55Short = 38

Color Constancy in complex

images




Observer Matches



Appearance = Reflectance ?

• Yes

Reflectance predicts appearance for flat Mondrians and the Munsell Book of Colors,

in uniform illumination.

McCann, McKee, and Taylor, Vision Res, 1976


Reflectance is defined as the ratio of

incident flux on a sample surface to reflected flux from the surface

Reflectance


Reflectance is defined as the ratio of

incident flux on a sample surface to reflected flux from the surface

No one measuresReflectance


Practical measurementsReflectance relative to a standard

Photocell

Object Illumination

LensStep 1: Radiance from Object


People measureReflectance relative to a standard

Lens PhotocellStep 2: Radiance from White

Object Illumination


Object Illumination

Object Illumination

LensRadiance from ObjectRadiance from White

Relative Reflectance =


Retinex Replace temporal with spatial

Calculate relative reflectance

by spatial comparisons

Lx,y / Lx*,y*

Lx,y

Lx*,y*

E. H. Land & J. J. McCann “Lightness and Retinex Theory”, J. Opt. Soc. Am. 61 1-11, 1971.


X, Y, Z pixel color

color

X separation

Y separation

Z separation

Pixel-based Colorimetry

Spatial-comparison Retinex


Auto-normalize with reset to maximaTuesday, October 20, 2009



LDR

HDR




LDR HDRscenes

11 reflectances (paints)2 identical sets of blocks

>200 facetsTuesday, October 20, 2009

uniformitygoal

LDR HDRilluminations

2 directionalsources


The world’s only gray world

LDR HDRillumination


LDR HDRscenes





Does illumination affect appearance?

Does appearance always correlate with

reflectance?


LDR

HDR

Multiple Exposures (x2)


LDR

HDR



LDR

HDR


14th October 2008Workshop 1 : Camera Capture and HDR WorkshopJohn McCann, Alessandro Rizzi, Carinna Parraman

(John McCann Imaging; Università Degli Studi Di Milano; CFPR, UWE)


Select Areas - LDR

GroundTruth


Select Areas - HDR


Select Areas - LDR


Select Areas - HDR


11 Paints


MLAB distance from constancy

MLAB coordinates

Munsell Chip Estimate of Appearance

Observed changes in Appearance

Munsell Notation of 11 paints


LDR / HDR Departures of Appearance from Reflectance

MLAB isotropic color spaceTuesday, October 20, 2009

LDR / HDR Departures 2



• Yes

• Most of the time

• Rarely


Two Questions about Colour Constancy?

• Does an object’s appearance = reflectance ?

• Only in perfectly uniform illumination

• Small changes observed in LDR

• Large changes in HDR

• Are appearances constant for all illuminants ?

• No

• Large changes in lightness, hue and chroma




Appearance = ?????

•Appearance ≠ Reflectance•Multiple reflections

•White reflectance•Appears white•Appears cool gray•Appears magenta•Appears yellow



LDR

HDR




Appearance = ?????

•Appearance ≠ Reflectance•Illumination edges

•Gray reflectance•Appears light gray•Appears dark gray•Appears black

• Illumination changes order of appearances


Measure LDR & HDR appearancesby painting with watercolors


Two Watercolor Renditions LDR HDRscene scene


BluePaint

4

LDR Painting Reflectance

0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

4 LDR

HDR Painting Reflectance

0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

4 HDR

L*=116R0.33-16Tuesday, October 20, 2009

BluePaint

4 & 29


0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

4 LDR 29 LDR


0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

4 HDR 29 HDR


BluePaint4 & 5

29


0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

4 LDR 5 LDR 29 LDR


0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

4 HDR 5 HDR 29 HDR


Watecolor Reflectances in LDR

0

20

40

60

80

100

400 500 600 700Wavelength (nm)

Lig

htn

ess

(L*

)

7 8 11

Watecolor Reflectances in HDR

0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

7 8 11

GreenPaint


RedPaint


0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

12 LDR 22 LDR 35 LDR


0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

12 HDR 22 HDR 35 HDR


Watecolor Reflectances in LDR

0

20

40

60

80

100


Lig

htn

ess

(L*

)

19 21 25

Watecolor Reflectances in HDR

0

20

40

60

80

100


Lig

htn

ess

(L*

)

4H 5H 29H

MagentaPaint



0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

30 LDR 32 LDR 33 LDR

18 LDR 9LDR


0

20

40

60

80

100

400 500 600 700

Wavelength (nm)

Lig

htn

ess

(L*

)

30 HDR 32 HDR 33 HDR

18 HDR 9 LDR

LDR HDR






• Yes

• Most of the time

• Rarely


Do humans discount illumination?

NoTuesday, October 20, 2009

Colour Constancy: 3 Types of Models

Model CalculationGoal

[Output]

Given Information

[Input]

Reference

Retinex appearanceradiance arrayof entire scene Land, 1971


CIELABCIECAM

appearancepixel’s radiance

+ pixel’s irradiance

CIE, 1976CIE 1997,2002

Computer Vision

reflectance radiance arrayof entire scene

Ebner, Funt, Finlayson,

Drew, 1998


Colour Constancy: 3 Types of Models

ModelCalculation

Goal[Output]

Mechanism [Output] = Reflectance

Retinex appearance build appearance from edges

sometimes


CIELABCIECAM

appearance measure reflectancestretch

always

Computer Vision

reflectanceestimate illumination

to calculate reflectance

always


Separate illumination from reflectance?


Vassillos Image Processing


Vassillos Image Processing



Color Reproduction

• 3-D Color Space [One into Another]


Colorant Systems


Interpolate Primaries

3-D Graphics, Bob Sobol, HP EmeritasTuesday, October 20, 2009

But, Real Primaries


But, Real Primaries75%R * 75%G * 75%B

75% * 75% * 75% = 42.2%


Color Reproduction

• 3-D Color Space [One into Another]

• Moving all the rooms in your house

• Into a smaller house*

* Why is the reproduction space smaller?


*Smaller space?

• Colorants - Physical spectra

• Gamut limits

• Unwanted absorptions

• Tone scale [Contone, Halftone, Transfer]

• Media [Print and display technologies]

• Surface properties limit range

• Veiling glare limits range


Outline



• Practice


• Gamut Mapping

• Theory


How the Print represents the WorldLightin theWorld

Film &Camera

• Reproducing paintings should be easy

• AgX photography

• Both are reflective

• Similar gamuts

• Expect good results

• Minimum effortTuesday, October 20, 2009

Lightness

0

20

40

60

80

100

0 20 40 60 80 100O r i g i n a l

Pri

nt Chemical

DigitalSlope 1.0

Black

White


Lightness

0

20

40

60

80

100

0 20 40 60 80 100O r i g i n a l

Pri

nt Chemical

DigitalSlope 1.0

Black

White

Expand

Compress

Compress

Slope = 1.0

CompressCompress

Slope = 1.0

Original Input

PrintExpand

White Black


Optical Density

0.00

0.50

1.00

1.50

2.00

2.50

0.00 0.50 1.00 1.50 2.00 2.50O r i g i n a l

Pri

nt Chemical

DigitalSlope 1.0

Black

White


S-curve enhances colorOptical Density

O r i g i n a l

Pri

nt

0.00

0.50

1.00

1.50

2.00

2.50

0.00 0.50 1.00 1.50 2.00 2.50

Chemical

Slope 1.0

Black

White

Optical Density

O r i g i n a l

Pri

nt

0.00

0.50

1.00

1.50

2.00

2.50

0.00 0.50 1.00 1.50 2.00 2.50

Chemical

Slope 1.0

Black

White

Optical Density

O r i g i n a l

Pri

nt

0.00

0.50

1.00

1.50

2.00

2.50

0.00 0.50 1.00 1.50 2.00 2.50

Chemical

Slope 1.0

Black

White

.05 in

0.6out

1.25 in

1.80 out

1.25 in

1.80 out

Optical Density

O r i g i n a l

Pri

nt

0.00

0.50

1.00

1.50

2.00

2.50

0.00 0.50 1.00 1.50 2.00 2.50

Chemical

Slope 1.0

Black

White

0.5 1.25 1.25

In Out

0.6 1.80 1.80


The story of how Replicas are made starts with a trip to the Museum. This is the main entrance to the Museum of Fine Arts, Boston. In addition, Replicas in this exhibit were made from photographs taken at the National Gallery, Washington and the Pushkin Museum, Moscow.

Polaroid Replicas


The Replica photographer uses an 8x10, or larger camera with a special light source to uniformly illuminate the painting.


In order to generate the feeling of depth the light must be both uniform and come from above the painting.

Look at the shadow cast by the light above the painting. The eye reads that shadow and makes the paint look “Three Dimensional.”


In order to record enough information about the actual colors in the painting, the photographer first makes a photograph of this 512 color calibration target. He places the target in front of the painting in the same light that will be used to photograph the painting.


This is a photograph of the Monetʼs “ Field of Poppies near Giverney” from the White Fund, housed at the Museum of Fine Arts, Boston.


This is the optical head of the digital drum scanner that records the image as digits. The scanner divides the photograph into 100 million picture elements--pixels. Each square pixel is 1/1200 of an inch on a side. Red, green and blue records have 256 levels -- 8 bits. We converted the photographic image to a 300 megabyte digital file.


This diagram shows how the each calibration color patch in the test target was created by a number (ie, 255,50,50) sent to a film recorder.

Start Here

DigitTriplet

255,50,50

Red DigitTriplet

239, 62,28

ApplyTransform

FilmTarget

Printer

Red

FilmPhoto of Painting & Target

CalculateTransform

Camera Scanner

“Master”Positive Tranparency

The scan of photo produces a new number (239,62,28). The calibration program calculates the effects of film, illumination, and processing. The transform board stores the calibration instructions and uses them to convert scanner digits to new output digits that remove color distortion.


We now convert the 300 megabyte digital file back to a new 'Master' photographic transparency. The FIRE 1000 film recorder writes a new 8x10 positive transparency (1200 dpi) that contains all the corrections from the calibration. It anticipates the characteristic of the print film so that the final image looks exactly like the original painting.


This image does not look like the original painting. It has lower contrast, exaggerated color saturation. The result is that the final print has a slope one relationship with the painting and exact color match. The distortions in the second transparency remove the characteristic properties of the print film.


This diagram shows the giant Polaroid processor in the room-size camera.

The light sensitive film--negative is at the top. The positive print paper is at the bottom. The chemical reagent, or developer, is applied between the positive and the negative. After the shutter opens, the strobe behind the transparent master discharges, and the shutter closes, the operator turns on a motor that drives a pair of titanium rolls that marries the positive and negative sheets and meters a precise amount of reagent between the sheets. The operator turns on the lights inside the camera, while the combine positive and negative are held in the air.

Role of photosensitive material-negative

Reagent Layer

Lens

Role of positve print paper

Film in exposure position

PressureRolls


This is a photograph of the outside of the room-sized camera.


The Replica photograph is held in the air with a line and pulley.

The print paper is white and the negative is black. Both are opaque.

The technicians turn on the lights, place the print on the floor, and wait 90 seconds for the dyes to transfer from the negative to the positive.


Two technicians peel away the black-backed negative leaving a finished actual-size reproduction of the painting.


The photograph is mounted on an archival board, spayed with a matte surface UV absorbing material. It is the placed in a hand-crafted frame for display.


Color PrintingCalibration


Make a regular digital arrayDigits

192,128,64



Closed Loop Calibration

Digits192,128,64

In

Digitsr,gb

Out 3D LUT

Digits192,128,64


3D LUT[ LookUp Table ]

R -

B - |

G

R G Bin

Find Nearest Neighbors1R G B = 1DENR DENG DENB2R G B = 2DENR DENG DENB3R G B = 1DENR DENG DENB4R G B = 3DENR DENG DENB5R G B = 4DENR DENG DENB6R G B = 5DENR DENG DENB7R G B = 6DENR DENG DENB7R G B = 7DENR DENG DENB8R G B = 8DENR DENG DENB

DENR DENG DENB

out

Interpolate


Film

Recorder

Digits192,128,64

Digits222,99,105

3D LUT

print

GraphicsArt

Scanner

transparency print

8x10View

Camera

print

Enlarger

transparency

MFA Monet original


Film

Recorder

Digits192,128,64

Digits222,99,105

3D LUT

print

GraphicsArt

Scanner

transparency print

Enlarger

transparency

MFA Monet original


MFA Monet original Replica

=


3D LUTP.C. Pugsley, Image reproduction

methods and Apparatus, British patent, 1,369,702(1974).

Kang, Color Technology of the Electronic Imaging Device


Printer with 3D LUTM. Abdulwahab, J. L. Burkhardt and J. J. McCann,

Method and Apparatus for Transforming Color Image Data on the Basis of an Isotropic and Uniform

Colorimetric Space, U. S. Patent, 4,839,721, Jun.13,1989


Kotera et. al.K.Kanamori, H.Kawakami, and H.Kotera:"A Novel Color

Transformation Algorithm and Its Applications", Proc. SPIE., vol.1244,p.272-281(1990) ;

K.Kanamori and H.Kotera:"Color Correction Technique for Hard Copies by 4-Neighbors Interpolation Method", Jour.Imag.Sci.&Tech., 36,1, p.73-80(1992);

K.Kanamori, H.Kotera, O.Yamada, H.Motomura, R.Iikawa, T.Fumoto: "Fast Color Processor with Programmable Interpolation by Small Memory(PRISM)", Jour.Electronic Imaging, vol. 2(3), pp.213-224(1993);

H.Kotera,K.Kanamori,T.Fumoto,O.Yamada,H.Motomura and M.Inoue:"A Single Chip Color Processor for Device Independent Color Reproduction",Proc. 1 st CIC, p.133-137(1993);

T.Fumoto, K.Kanamori, O.Yamada, H.Motomura, and H.Kotera : "SLANT/PRISM Convertible Structured Color Processor MN5515, Proc. 3rd CIC, p.101-105(1995).


Outline



• Practice


• Excellent results - 3D LUTS

• Each “room” separately

• See approach in device profiles


Observations

• A. Color reproduction is like moving into a smaller house

• We cannot just shrink everything to fit

• Different solution for each room*

• Color

• 3-D lut [very powerful and affordable]

• Best reproductions preserve edges


Outline



• Practice


• Gamut Mapping

• Theory


Color Display Gamut


SWOP (coated) Gamut


Original / Reproduction

Perfect

A Ar BrB

Ar = A Br = B

Ar / Br = A / B

89 95 89 95


Conserverve X,Y,Z

89 95 89

85

A Ar BrB

Ar = A Br = .9B

Ar / Br ≠ A / B

85

Out of gamutSubstitute


89 95

Out of gamutSubstitutes

Conserve Spatial Ratio A Ar BrB

80 85

Ar = .9A Br = .9B

Ar/Br = A / BTuesday, October 20, 2009

L*a*b* & Color Appearance

Original


Copy A Copy B

∆ECopy A = ∆ECopy BTuesday, October 20, 2009

Original Copy A


Original Copy B


17 Areas

∆ECopy A = ∆ECopy B

∆E ∆EArea [Or ig&CopyA] [Or ig&CopyB]

1 17 .3 17 .02 17 .0 17 .23 15 .2 17 .94 17 .3 17 .35 18 .5 17 .56 16 .2 18 .07 17 .9 15 .78 17 .3 17 .59 17 .3 18 .0

1 0 17 .3 17 .51 1 17 .3 17 .31 2 17 .3 18 .01 3 17 .3 17 .51 4 17 .3 16 .51 5 17 .3 18 .01 6 17 .3 17 .91 7 17 .3 17 .3

Average 17 .2 17 .4


32 SpatialEdge Ratios

used in calculating

ΔR

Single pixel used in

calaculating ΔE


Original Copy A


Original Copy B


Lab Copy

X’Y’Z’

XYZ

Copy

XYZ

X’Y’Z’

Original XO/X’ O

XC/X’ C

2)YC/ Y′CYO/ Y′O(1- +√ΔR= )2(1- ZC/ Z′C

ZO/ Z′O+2)XC/ X′C

XO/ X′O(1-

Lab Original

√ΔE= +)2(aC- aO+L C- L O)2( bC- bO)2(

ΔR compares RatiosΔE compares Pixels


L*a*b*pixel

Lightin theWorld

Retinexfield

=


5 R5 YR5 Y5 GY5 G5 BG5 B5 PB5 P5 RP

Munsell

7/6


Conserve Spatial Ratio Lightness

OriginalOriginal+ 1 M chip

Original- 1 M chip


Conserve Spatial Ratio Chroma

OriginalOriginal- 1 M chip

Original+ 1 M chip


Conserve Spatial Ratio Hue


Original+ 1 M chip


Alter Spatial Ratio Random


Original+ 1 M chip


Calculating the best extra-gamut colors

using spatial comparisons

[Retinex]


The Experiment

BestIn . tiff

GoalIn, tiff


Toyo Inks (uncoated)

BestIN . tiff

GAMUTWARNING


1. Create Multi-resolution BestIn


2. Create Multi-resolution GoalIn


Retinex Color Gamut


3. Separate r, g, b channels

create r, g, b color-separation channels


-create r, g, b color-separation channels



Result

BestIn . tiff GoalIn, tiff

BOUT.tiff

In-gamut output

looks like Goal


GOAL BEST

SWOP coated


GOAL BEST

SWOP coated


LUT


LUT


GOAL

BEST GamutRetinex


GOAL

BEST

GamutRetinex


Lessons

Searching for matching color appearances using spatial comparisons (Retinex) finds in-gamut images that do look like the Goal.

Reason: Gamut Retinex uses the spatial relationships that control human color appearance.

It uses spatial comparisons as input from the Goal image.


LessonsColor matches (CIE) for extra-gamut colors

do not look like the Goal image colors.

Reason

Color matches (CIE) are calculated one pixel at a time.

The relationship to other pixels is ignored.

The CIE process distorts the spatial relationships that control human color appearance.


Corollary 1 The only time that CIE color matches will

make a reproduction look exactly like the original is when all the individual pixels match.

Reason: Any extra-gamut pixel will introduce new spatial relationships with the rest of the image that are different from those of the original.

These relationships can distort the appearance of the entire image.


Corollary 2 When spatial comparisons (Retinex) finds in-gamut images that do look like the Goal, it will significantly

increase the deltaE of corresponding pixels.

Reason: In order to get the correct relationships throughout the image the calculation will increase delta E for in-gamut pixels.


Observations




• Color



• B. Color gamut mapping is like Color reproduction





• Color

• Different optimization for each room



• B. Color gamut mapping is like Color reproduction

• C. Best done in an isotropic color space [UCS]

Observations


Outline



• Practice


• Gamut Mapping

• Theory


Theory

• Colorant Primaries

• Interpolate in 3-D color spaces


Colorant Systems


Subtractive


Subtractive


Subtractive


Newsprint


Halftone


Theory• Colorant Primaries• Interpolate in 3-D color spaces

• [Which one ??????]• Should be visually isotropic• e.g. We can average data• Uniform Color Space [UCS]

• Munsell Space• LMS Cones• XYZ• CIELAB• CIECAM


• Uniform Color Space [UCS]• Munsell Space

Dorothy Nickerson, 1943


Spectral Sensitivity

Maxwell 1860

CIE 1931Brown Wald 1970

Color Theory

Data



BROWN & WALD’SVISUAL PIGMENTS

• LMS Cones


WRIGHT’SCOLOR MATCHES

CIE 1931

X, Y, Z space


L*p =[ 116 (Y/Yn)1/3 -16 ]a*p= 500 [(X/Xn)1/3-(Y/Yn)1/3 ]b*p= 200 [(Y/Yn)1/3-(Z/Zn)1/3 ]

L*pa*pb*p =[ f(Xp,Yp,Zp) ]

L*a*b* is the CIE 1976 Space and Color Difference Formula


L*p =[ 116 (Y/Yn)1/3 -16 ]a*p= 500 [(X/Xn)1/3-(Y/Yn)1/3 ]b*p= 200 [(Y/Yn)1/3-(Z/Zn)1/3 ]

L*pa*pb*p =[ f(Xp,Yp,Zp) ]



L*p = Yrefl1/3 a*p = 500 [ Xrefl1/3 - Yrefl1/3 ]b*p = 200 [ Yrefl1/3 - Zrefl 1/3 ]



Lightness ?????Receptor PhysiologyResponse = Log (Intensity light)

Psychophysical ResultLightness = (Intensity light) 1/3

Scattered light in eyeball


Lightness = (Intensity light) 1/3

Retina = Log (Intensity light)


L*p = YreflAS a*p= 500 * [ XreflAS - YreflAS ]b*p= 200 * [ YreflAS - ZreflAS]



L*a*b* / CIE 1976 Space and Color

Difference Formula

500 x Stretch

• LMS Cones



Summary• Scene• Scene dependent intraocular scatter (spatial)

• Limits HDR• Scene dependent appearance (spatial)• Best reproduction matches appearance

• Smaller color spaces prevent matching all pixels• Calculation intensive • Best done in Uniform Color Space (which one?)

• Practice• Reproduce Paintings (one room at a time)• Gamut Mapping (spatial - preseve edges)


Conservation

• Planet• Environment • Resources• Energy


Conservation

• Planet• Environment • Resources• Energy

• Fine art


Conservation • Planet

• Environment • Resources• Energy

• Fine art

• Edges in reproductions


[email protected]@[email protected]://web.mac.com/mccanns/McCannImaging/Home.html

Thank you


mailto:[email protected]






http://web.mac.com/mccanns/McCannImaging/Home.html

http://web.mac.com/mccanns/McCannImaging/Home.html

Color Reproduction / Color Gamut Mapping: Two Sides of - Create!

Documents