framework for printing with daylight fluorescent inks · 2018. 7. 17. · red, green and blue srgb...

POUR L'OBTENTION DU GRADE DE DOCTEUR ÈS SCIENCES

acceptée sur proposition du jury:

Prof. M. Pauly, président du juryProf. R. Hersch, directeur de thèse

L. Choulet, rapporteur Prof. S. Süsstrunk, rapporteur

Dr Ph. Urban, rapporteur

Framework for printing with daylight fluorescent inks

THÈSE NO 5636 (2013)

ÉCOLE POLYTECHNIQUE FÉDÉRALE DE LAUSANNE

PRÉSENTÉE LE 18 JANVIER 2013

À LA FACULTÉ INFORMATIQUE ET COMMUNICATIONSLABORATOIRE DE SYSTÈMES PÉRIPHÉRIQUES

PROGRAMME DOCTORAL EN INFORMATIQUE, COMMUNICATIONS ET INFORMATION

Suisse2013

PAR

Romain ROSSIER

The only laws of matter are those that our minds must fabricate and the only laws of mind are

fabricated for it by matter.

— James Clerk Maxwell

To my wife and my parents for their endless support and encouragement

AcknowledgementsFirst, I have to thank my wife Marie. I met her at the beginning of my undergraduate studies.

During many years she gave me an endless support and encouragement. She always believed

in me giving me the strength to reach my goals. I also thank my parents, my brothers and my

sister who always encouraged me.

Thank you Roger for having given me the opportunity of completing a Phd at the EPFL. You

provided many challenging and interesting research topics. Thank you for your support, the

weekly meetings and the discussions about my work and progress. You were sincerely involved

in my different projects. Both your assistance and your guidance are invaluable.

I also express my gratitude to Julien Andres for all the discussions about my questions related

to chemistry. A special thank to Thomas Bugnon who was senior Phd student when I arrived

at the LSP laboratory and kindly always found time to answer my questions. I also express my

gratitude to Sergiu Gaman who programmed under my guidance useful applications and to

Pascal Fehr who designed some images used in my thesis.

During my Phd studies I had good times with friendly people. The excellent atmosphere at my

working place contributed to make my studies enjoyable. Thank you Xavier Jimenez, Mathieu

Hébert, Vahid Babaei, Petar Pjanic, Basile Schaeli, Florent Garcin, Mathieu Brichon, Sebastian

Gerlach, Fabienne Allaire, Maria Anitua, Andrea Maesani, Jean-Marc Comby, Fabrice Rousselle,

Zhe Wei, Rafik Chaabouni, Marjan Shahpaski and Sylvain Chosson.

I also express my gratitude to the Swiss National Foundation for its financial support, grant n0

200020-126757/1.

Lausanne, 27 Juin 2012 R. R.

v

AbstractDuring the last century technical aspects of printers such as resolution, fastness and quality of

the prints have been steadily improved. It is now common to have at home desktop printers

enabling quickly printing images with a good reproduction quality. However, since most of

the printers use the three cyan, magenta and yellow primaries complemented with the black

ink, the domain of printable color has not been significantly improved. The domain of color is

limited to the color produced according to the combinations of classical cmyk inks.

In this thesis, we are interested in adding an extra dimension to color prints by making use of

the daylight fluorescent magenta and yellow inks. By combining these fluorescent inks with

classical cmyk inks, we obtain high chroma and bright colors. We propose a new approach for

expanding the gamut of classical cmyk printers by using these fluorescent inks with classical

cmyk inks in a 6 ink print setup. The approach comprises first establishing a spectral prediction

model dedicated to the accurate spectral prediction of halftones comprising classical and

daylight fluorescent inks. This new spectral prediction model, although calibrated with a few

halftone patch reflectances, is remarkably accurate for predicting spectral reflectances of both

classical ink and combined classical ink and daylight fluorescent ink halftones printed with

offset and inkjet printers. It also shows excellent prediction accuracies for classical halftones

printed with electrophotographic printers.

The approach also comprises a gamut mapping of the sRGB display gamut to the fluorescent

printer gamut which has colors beyond the sRGB display gamut. The goal is to enhance image

parts by printing them with high chroma and bright daylight fluorescent colors. We first select

the image parts to be enhanced. We then apply to their colors a gamut expansion that increases

both their chroma and their lightness towards the colors located at the boundary of the gamut

formed by the combination of classical and daylight fluorescent inks. This expansion can be

controlled by user-defined parameters. We create smooth chroma transitions between the

expanded and non-expanded image parts. We also preview the printable gamut expanded

image generated according to user-defined gamut expansion parameters. The resulting

prototype software enables artists to create and print their own designs.

Gamut expansion parameters can be set to limit printable colors to sRGB gamut colors. In this

situation, we are taking advantage of the larger fluorescent printer gamut without considering

colors located beyond sRGB gamut colors. By having both a larger printer gamut and a suitable

gamut mapping from the sRGB gamut to the fluorescent printer gamut, we are able to better

reproduce images such as images of watches and master paintings which are known to have

colors outside the gamut of classical inks.

vii

Acknowledgements

By characterizing the fluorescent printer gamut, we show that a significant part of the classical

cmyk printer gamut can be reproduced by combining classical and daylight fluorescent inks.

By printing parts of images with a combination of classical and daylight fluorescent inks

instead of using classical inks only, we can hide security patterns within printed images. Under

normal daylight, we do not see any difference between the parts printed with classical inks

only and the parts printed with daylight fluorescent inks and classical inks. By changing

the illumination, e.g. by viewing the printed image under a tungsten lamp or UV lamp,

the daylight fluorescent inks change their color and reveal the security patterns formed by

combinations of classical and daylight fluorescent inks. We also show how to hide security

patterns under at the same time various natural and artificial illuminations. These security

patterns are revealed under an illumination having energy only in the excitation wavelengths

of the daylight fluorescent inks, such as a UV or non UV blue illumination.

Keywords : color prints, color reproduction, spectral prediction models, daylight fluorescent

inks, gamut mapping, gamut expansion, optical document security.

viii

RésuméAu court du dernier siècle, les aspects techniques des imprimantes tels que la résolution, la

rapidité et la qualité des imprimés ont étés régulièrement améliorés. Il est de nos jours tout

à fait commun d’avoir à la maison des imprimantes de bureau qui permettent d’imprimer

rapidement des images avec une bonne qualité de reproduction. Toutefois, comme la plupart

des imprimantes utilisent les trois primaires cyan, magenta et jaune complémentées par le

noir, le domaine imprimable de couleur n’a pas été significativement amélioré. Ce domaine

de couleur est limité aux couleurs produites par les encres classiques CMJN.

Dans cette thèse, nous investiguons la possibilité d’ajouter une dimension supplémentaire

aux imprimés couleurs par l’utilisation des encres magenta et jaunes fluorescentes à la lu-

mière visible. En combinant ces encres fluorescentes avec les encres classiques CMJN, nous

obtenons des couleurs plus lumineuses et de chroma élevé. Nous proposons une nouvelle

approche d’expansion de la gamme de couleur des imprimantes CMJN classiques par l’utilisa-

tion d’encres fluorescentes à la lumière visible avec des encres classiques dans un système

d’impression à 6 encres. Cette approche comprend premièrement l’établissement d’un modèle

de prédiction spectral dédié aux prédictions précises de demi-tons comprenant des encres

classiques et fluorescentes à la lumière visible. Ce nouveau modèle de prédiction spectral,

bien que calibré avec peu de réflectances de demi-tons, est remarquablement précis lors de la

prédiction de réflectances spectrales de demi-tons comprenant à la fois des encres classiques

et la combinaison d’encres classiques avec des encres fluorescentes à la lumière visible lors

d’impressions avec des imprimantes jet d’encre et offset. Il se montre aussi très précis pour la

prédiction de demi-tons classiques imprimés avec des imprimantes électrophotographiques.

Cette approche comprend aussi un mappage du gamut des écrans sRGB vers le gamut fluores-

cent de l’imprimante qui a des couleurs au-delà de la gamme de couleurs des écrans sRGB. Le

but est ici d’améliorer des parties d’images en imprimant celles-ci avec le fort chroma et la

puissante luminosité des couleurs fluorescentes à la lumière visible. Premièrement, nous sélec-

tionnons les parties d’images à être améliorées. Nous appliquons ensuite à leurs couleurs une

expansion de la gamme des couleurs qui augmente à la fois leur chroma et leur luminosité en

direction des couleurs positionnées à la frontière du gamut formé par la combinaison d’encres

classiques et fluorescentes à la lumière visible. Cette expansion peut être contrôlée par des

paramètres définis par des utilisateurs. Nous créons des transitions de couleur douces entre

les parties d’images étendues et non étendues. Nous fournissons aussi un aperçu à l’écran

de l’image étendue selon les paramètres d’expansion définis par l’utilisateur. Le prototype de

logiciel résultant permet aux artistes de créer et d’imprimer leurs propres conceptions.

ix

Acknowledgements

Les paramètres d’expansion de la gamme de couleur peuvent être définis de manière à limiter

les couleurs imprimables aux couleurs affichables par un écran sRGB. Dans ce cas, nous tirons

avantage de la large gamme de couleur fluorescente de l’imprimante sans considérer les

couleurs au-delà de la gamme de couleur sRGB. En ayant à la fois une gamme de couleur

d’imprimante étendue et un mappage des couleurs approprié du gamut sRGB vers le gamut

fluorescent de l’imprimante, nous sommes capables de mieux reproduire les images, telles

que des images de montres et de peintures qui sont connues pour avoir des couleurs au-delà

des couleurs reproductibles par les encres classiques CMJN.

En caractérisant le gamut fluorescent de l’imprimante, nous montrons qu’une partie signi-

ficative du gamut d’imprimantes CMJN classiques peut être reproduite en combinant des

encres classiques et fluorescentes à la lumière visible. En imprimant des parties d’image avec

une combinaison d’encres classiques et d’encres fluorescentes à la lumière visible à la place

d’utiliser des encres classiques seulement, nous pouvons cacher des motifs de sécurité dans

des images imprimées. Dans des conditions normales de lumière du jour, nous ne percevons

aucune différence entre les parties imprimées avec des encres classiques seulement et les par-

ties imprimées avec des encres classiques et fluorescentes à la lumière visible. En changeant

l’illumination, par exemple en regardant l’image imprimée sous une lampe tungstène ou une

lampe UV, les encres fluorescentes à la lumière visible change leur couleur et révèlent les

motifs de sécurité formés par combinaisons d’encres classiques et fluorescentes à la lumière

visible. Nous montrons aussi comment cacher à la fois sous plusieurs lumières artificielles

et naturelles des motifs de sécurité. Ceux-ci sont révélés sous une lumière ayant de l’énergie

seulement dans les longueurs d’onde d’excitation des encres fluorescentes à la lumière visible.

Mots-clefs : imprimés en couleur, reproduction couleur, modèles de prédictions couleur,

encres fluorescentes à la lumière visible, expansion de gamut, sécurité optique de documents.

x

Contents

Acknowledgements v

Abstract (English/Français) vii

List of figures xii

List of tables xv

1 Introduction 1

1.1 Motivations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Challenges . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2

1.3 Characteristics of daylight fluorescent colorants . . . . . . . . . . . . . . . . . . . 3

1.4 Dissertation outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

1.5 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

1.6 Review of the prior art . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10

2 Prediction model for classical and daylight fluorescent inks 13

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2 Limitation of the Yule-Nielsen model for predicting fluorescent ink halftones . 13

2.3 The ink spreading enhanced cellular Yule-Nielsen model . . . . . . . . . . . . . 14

2.3.1 Ink spreading extension of the Cellular Yule-Nielsen model . . . . . . . . 14

2.4 Characterizing ink spreading with sensor responses . . . . . . . . . . . . . . . . 19

2.5 Prediction accuracies for classical ink halftones . . . . . . . . . . . . . . . . . . . 21

2.6 Prediction accuracies for fluorescent ink halftones . . . . . . . . . . . . . . . . . 22

2.7 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Framework for printing with combined classical and daylight fluorescent inks 27

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

3.2 Comparison of fluorescent and non-fluorescent ink gamuts . . . . . . . . . . . . 28

3.2.1 Comparison of gamut volumes . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.3 Comparison of inkjet and offset fluorescent ink gamuts . . . . . . . . . . . . . . 33

3.4 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

xi

Contents

4 Gamut mapping expansion and reduction for color reproduction with daylight fluo-

rescent inks 35

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

4.2 Mapping the lightness range of the sRGB gamut into the ink destination gamut 35

4.3 Mapping the lightness adapted sRGB gamut onto the printable fluorescent gamut 38

4.4 User driven gamut expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

4.5 Display preview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.6 Halftoning and printing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

4.7 Summary of application user-defined parameters . . . . . . . . . . . . . . . . . . 47

4.8 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

5 Gamut expanded images 49

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

5.2 Preview and corresponding print of a gamut expanded image . . . . . . . . . . . 49

5.3 Advertising images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

5.4 Better reproduction of input sRGB image colors . . . . . . . . . . . . . . . . . . . 54

5.5 Artistic images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

5.6 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

6 Hiding patterns with daylight fluorescent inks 61

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

6.2 Hiding security patterns by printing colors either with or without daylight fluo-

rescent inks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62

6.2.1 Illustrations of hidden security patterns . . . . . . . . . . . . . . . . . . . 65

6.3 Hiding a variable intensity security image . . . . . . . . . . . . . . . . . . . . . . 66

6.4 Hiding security patterns under multiple illuminations . . . . . . . . . . . . . . . 69

6.5 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

7 Conclusion 73

Bibliography 80

A Hue planes of the fluorescent gamuts of both an offset and an inkjet printer 81

B Curriculum Vitæ 85

xii

List of Figures1.1 Spectral power distribution of the F7, D65 and A illuminants . . . . . . . . . . . 4

1.2 Reflectance factors of inkjet daylight fluorescent colorants . . . . . . . . . . . . 5

1.3 Comparison between inkjet and offset daylight fluorescent colorants . . . . . . 6

1.4 Fading of the inkjet and offset daylight fluorescent m f and y f colorants . . . . 8

2.1 Illustration of one of the 8 subdomains of the cellular Yule-Nielsen model . . . 15

2.2 Cyan dot gain curve for a cmy laser print . . . . . . . . . . . . . . . . . . . . . . . 18

2.3 Cellular Yule-Nielsen model accounting for ink spreading . . . . . . . . . . . . . 19

3.1 Comparison between the 3D Gcmyk , G f and lightness adapted G ′sRGB gamuts . 29

3.2 Color gamuts of the non-linearly lightness adapted sRGB space, the classical

cmyk 4 ink print setup and the 6 ink print setup combining the cmyk inks with

the m f and y f inks under the D65 illuminant . . . . . . . . . . . . . . . . . . . . 30

3.3 Comparison of the non-linearly lightness adapted sRGB gamut and the fluores-

cent 6 ink gamut under the D65, F7 and A illuminants . . . . . . . . . . . . . . . 31

3.4 Comparison of the inkjet and offset fluorescent G f gamuts under the D65 illumi-

nant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.1 Linear and non-linear mapping of sRGB lightnesses . . . . . . . . . . . . . . . . 36

4.2 Comparison between the linearly and non-linearly lightness adapted display

gamut and the fluorescent ink gamut. . . . . . . . . . . . . . . . . . . . . . . . . . 37

4.3 Mutiple foci gamut mapping approach for both gamut expansion and reduction 38

4.4 Constant hue planes for an inkjet printer showing both constant and non-

constant lightness mapping lines at hue angles of the magenta and green fluo

colorants. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

4.5 Constant hue planes for an offset printer showing both constant and non-

constant lightness mapping lines at hue angles of the yellow and red fluo colorants. 41

4.6 Spatial interpolation map for an arbitrary selection generated with a limitation

factor κ= 2.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

4.7 Luminance gamma curves γlum for both the Dell U2212 HM and the Eizo Color-

Graphic CG245W displays . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

5.1 Photographs of previewed prints of a gamut expanded lizard image and corre-

sponding both classical cmyk and fluorescent prints. . . . . . . . . . . . . . . . . 50

xiii

List of Figures

5.2 Photographs of printed both non-gamut expanded and gamut expanded Rolex

Yachtmaster images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5.3 Fluorescent and non-fluorescent ink color separation layers of the Yatchamster

image shown in Figure 5.2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

5.4 Photographs of printed both non-gamut expanded and gamut expanded AMG

car images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

5.5 Photographs of printed both non-gamut expanded and gamut expanded lipstick

advertisement images. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

5.6 Photographs of a printed advertisement Hublot pink gold watch, printed with

the ECI ISO Coated V2 profile and the fluorescent ink gamut G f . . . . . . . . . . 55

5.7 Photographs of a printed advertisement Rolex yellow gold watch, printed with

the ECI ISO Coated V2 profile and the fluorescent ink gamut G f . . . . . . . . . . 56

5.8 Photographs of a Claude Monet master painting printed with the ECI ISO Coated

V2 profile and the fluorescent ink gamut G f . . . . . . . . . . . . . . . . . . . . . . 57

5.9 Photographs of a part of a J. M. W. Turner master painting printed with the ECI

ISO Coated V2 profile and the fluorescent ink gamut G f . . . . . . . . . . . . . . . 57

5.10 Photographs of a Paul Gauguin master painting printed with the ECI ISO Coated

V2 profile and the fluorescent ink gamut G f . . . . . . . . . . . . . . . . . . . . . . 58

5.11 Photographs of a flower image printed with the ECI ISO Coated V2 profile and

both non-gamut expanded and non-linearly gamut expanded and printed with

the fluorescent ink gamut G f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.12 Photographs of a designed flaming girl image printed with the ECI ISO Coated

V2 profile and of the same image linearly gamut expanded and printed with the

fluorescent offset gamut G f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59

5.13 Photographs of a fluorescent mushroom image printed with the classical inkjet

Gcmyk gamut and linearly gamut expanded and printed with the fluorescent ink

gamut G f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

6.1 Color gamut Gcmyk of the classical cmyk inkjet ink set and the strictly fluorescent

inkjet gamut Gs f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

6.2 Color gamut Gcmyk of the classical cmyk offset ink set and the strictly fluorescent

offset gamut Gs f . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

6.3 Example of an image design incorporating the hidden "VALID" security text. . . 65

6.4 Photographs of a printed Japanese girl image incorporating the repetitive "VALID"

pattern viewed under both daylight and UV illuminations. . . . . . . . . . . . . 66

6.5 Photographs of a printed Iceland landscape incorporating the repetitive "VALID"

pattern viewed under normal daylight, under UV illumination and under A

illumination. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

6.6 Halftoned variable intensity tiger image that is to be incorporated within a

security image. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

6.7 Photographs of a printed grayscale girl image incorporating a variable intensity

tiger image viewed under both normal daylight and under a UV illumination. . 68

xiv

List of Figures

6.8 Character L of the "VALID" mask message to be hidden under different illumi-

nants at 20% gray level intensity together with an enlargement of a small region

of its corresponding blue noise halftone . . . . . . . . . . . . . . . . . . . . . . . . 70

6.9 Photographs of an offset printed Iceland landscape incorporating the repetitive

"VALID" pattern viewed under normal daylight, under illumination A, under

fluorescent tube illumination F7 and under a blue low consumption Swiss light

classic 55 non UV lamp. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

6.10 Photographs of an inkjet printed Iceland landscape incorporating the repetitive

"VALID" pattern viewed under both normal daylight and UV illuminations with

different printed maximal daylight fluorescent cmy f ink surface coverages. . . 72

A.1 Comparison of constant hue planes between the fluorescent gamut of an inkjet

printer and the non-linearly lightness adapted sRGB gamut. . . . . . . . . . . . . 82

A.2 Comparison of constant hue planes between the fluorescent gamut of an offset

printer and the non-linearly lightness adapted sRGB gamut. . . . . . . . . . . . . 83

xv

List of Tables1.1 ∆E94 color difference between original offset and inkjet m f and y f measures

and after specific period of time under daylight behind a window glass . . . . . 7

2.1 Prediction accuracies of the IS-YNSN model for 125 cm f y f test samples printed

with an inkjet Epson P50 printer and an offset printer . . . . . . . . . . . . . . . 14

2.2 Prediction accuracies of both the IS-YNSN model and the IS-CYNSN models for

classical cmy inkjet and electrophotgraphic prints . . . . . . . . . . . . . . . . . 23

2.3 Prediction accuracies for both the IS-YNSN model and the IS-CYNSN models

when characterizing ink spreading using simulated RGB sensors. . . . . . . . . . 23

2.4 Prediction accuracies of the CYNSN, IS-YNSN and IS-CYNSN models for the

fluorescent and non-fluorescent inkjet ink sets used to establish the fluorescent

G f printer gamut. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

2.5 Prediction accuracies of the CYNSN, IS-YNSN and IS-CYNSN models for the

fluorescent and non-fluorescent offset ink sets used to establish the fluorescent

G f printer gamut. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

3.1 Prediction accuracies of the IS-CYNSN model for 625 cmyk and 125 cm f y f test

samples printed with a Canon Pro 9500 inkjet printer and measured under both

the A and F7 illuminants. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

3.2 Comparison of gamut volumes for the non-linearly lightness adapted sRGB gamut 32

4.1 Comparison of gamut volumes for the linearly lightness adapted sRGB gamut . 37

4.2 Prediction accuracies for both the Dell U2212 HM and the Eizo ColorGraphic

CG245W display characterizations . . . . . . . . . . . . . . . . . . . . . . . . . . . 45

xvii

1 Introduction

1.1 Motivations

In a society where images take an important place, they must satisfy high quality criteria.

Thus, they must be pleasant, faithful and their quality must be well controlled. Traditional

printing technologies, such as offset printing, electrophotgraphy printing and inkjet printing

use classical cmyk inks. In order to reproduce specific colors, e.g. brand colors, offset printers

can also use additional Pentone colors. However, most printing systems only perform color

separation with the classical cmyk inks.

Daylight fluorescent colorants were introduced in the middle of the last century, for devices

which should capture the attention, such as road markers, safety jackets and warning signs.

They were also used for graphic arts, mainly for painting and decoration. At present, they are

used in many products, such as highlighting markers, toys and optical brighteners in tissues

and papers. The main characteristic of daylight fluorescent colorants is that their colors are

brighter and more saturated than the corresponding classical colors. By combining these

colorants with classical colorants, one may increase the overall reproducible gamut.

In this thesis, we propose to create a color management framework by combining classical

cmyk inks with the daylight fluorescent magenta and yellow inks in a 6 ink printing system.

We would like to explore the possibilities offered by the high chroma and bright daylight

fluorescent inks. These daylight fluorescent inks enable improving the reproduction of color

images, especially in respect to bright and saturated colors. Since the daylight fluorescent

colorants are much brighter and more saturated than the classical colorants it is also possible

to highlight image regions of special interest.

We also would like to create specific tools for exploiting the capabilities of the fluorescent 6 ink

printing system. These tools may enable selecting image regions, applying to specific regions

different gamut expansions reinforcing chroma and brightness in these regions, displaying a

preview of the enhanced image having colors located beyond sRGB gamut colors and printing

the resulting gamut expanded image. Such an application offers new means to designers

1

Chapter 1. Introduction

working in fields such as photography, advertisement and magazine production.

Another application of fluorescence is the authentication of security documents. By printing

parts of images with a combination of classical and daylight fluorescent inks instead of using

classical inks only, we can hide security patterns within printed images. We are interested

in applying a metameric color match under a specific illumination, e.g. the D65 illuminant,

between the printed parts with combinations of classical inks and daylight fluorescent inks

and the printed part with classical inks only. By changing the illumination, e.g. by viewing

the image under a tungsten lamp or under a UV lamp, the daylight fluorescent inks change

their colors and reveal the security patterns formed by combinations of classical and daylight

fluorescent inks.

1.2 Challenges

In order to propose a comprehensive and complete solution to print images on a 6 ink printing

system using the classical cyan, magenta, yellow, black inks and the two additional daylight

fluorescent yellow and magenta inks, we need to characterize this printing system.

Characterizing a printer incorporating daylight fluorescent inks induces many difficulties.

First, classical spectral prediction models, such as the Yule-Nielsen modified spectral Neuge-

bauer model (Viggiano 1990) and the Clapper-Yule model (Clapper and Yule 1953) do not

predict well fluorescent ink halftones. In this situation, we have to develop a spectral predic-

tion model dedicated for predicting the spectral reflectance of halftones comprising daylight

fluorescent inks. Accurate spectral predicitons are needed for establishing the printable des-

tination fluorescent gamut as well as for creating a correspondence between CIELAB colors

and corresponding fluorescent ink dot surface coverages. In addition, when hiding security

patterns, an exact relationship between CIELAB colors and ink surface coverages enable

printing perfectly metameric colors between image parts printed either with combinations of

classical and daylight fluorescent inks or with classical inks only. The patterns will be therefore

perfectly hidden under a specific illumination, i.e. the illumination used for calibrating the

spectral prediction model. We also would like a spectral prediction model calibrated with a

few halftone patch reflectances.

In addition, daylight fluorescent inks strongly change in appearance when changing the

illumination, i.e. the total reflectance is strongly illuminant dependent. Depending on the

nature of the fluorescent inks, some fading effects may come up, when putting them under

the exposure of normal daylight (Connors-Rowe et al. 2005).

The problem of printing with custom inks raises similar problems as printing with combined

classical and fluorescent inks. There is a need for selecting specific subsets of inks from many

possible ink subsets, for mapping the input gamut into the gamut achievable with the multi-

ink halftones and for allowing the ink layers to be printed without inducing undesired moiré

layer superposition effects.

2

1.3. Characteristics of daylight fluorescent colorants

Producing gamut expanded images raises several challenges. We have to determine expansion

factors increasing the chroma of input sRGB colors to colors beyond the sRGB gamut and

possibly modify their lightnesses. The goal is to enhance given image parts with higher chroma

and brighter colors. We also have to ensure the continuity of colors at the boundary between

highlighted and non-highlighted image regions. In addition, we have to generate halftoned

images comprising at different locations different gamut mappings between the input image

and the destination image. In order to preview the printable expanded images having colors

beyond the sRGB gamut, we have to simulate a lower quality display for classical image parts

and render the extended sRGB colors by making use of the full capabilities of the display. Such

a preview enables visualizing the differences between the color expanded and non-expanded

image parts.

1.3 Characteristics of daylight fluorescent colorants

Daylight fluorescent inks contain organic molecules (Streitel 2009) that fluoresce by absorbing

light within one wavelength range and remitting light at a longer wavelength range. Classical

daylight fluorescent inks, such as the daylight fluorescent yellow ink and the daylight fluores-

cent magenta ink are mainly excited in the visible region between 400 and 550 nm. In addition,

they have an ultra-violet narrow excitation band between 350 and 400 nm (Connors-Rowe et al.

2005). The emission peak is located in the visible spectrum at wavelengths corresponding to

the desired color, i.e. for the daylight fluorescent magenta ink, the two emission peaks yield a

very bright and strongly saturated magenta color. These daylight fluorescent inks therefore do

not behave like classical inks where part of the incident light is absorbed by the inks. They

behave additively, i.e. the fluorescent emission behaves as a color light source.

The total reflectance factor Rtotal(λ) of a fluorescent ink patch is the light reflected by that

ink patch plus the light emitted by fluorescence divided by the light reflected by a perfect

Lambertian white reflector (Grum 1980). Both the UV and the visible range of an illuminant

have an impact on the energy that is available for fluorescent emission. In order to illus-

trate the strong impact in color appearance of daylight fluorescent colorants when changing

the illuminant, we consider the A, D65 and F7 illuminants. Figure 1.1 shows the spectral

power distribution of these illuminants. They have been measured with a Maya 2000 Pro

spectrophotometer calibrated from the known spectral power distribution of an Ocean Optics

DHL-2000-BAL lamp. The spectral power distribution of illuminant F7 has been measured

from a Just Normlicht mini 5000 light table. Both the A and the D65 illuminant emulations are

measured light sources of a SpectroEye Xrite spectrophotometer.

The color appearance of daylight fluorescent colorants is characterized by the total spectral

reflectance factor Rtotal(λ) under a given illuminant. From now on, we use the term reflectance

as a short denomination of total reflectance factor. Figure 1.2 provides the measured spectral

reflectance of four daylight fluorescent colorants, the daylight fluorescent m f magenta ink

(Farbel Castel ink ref. 154928), the daylight fluorescent y f yellow ink (Farbel Castel ink ref.

3


450 500 550 600 650400 700

1

2

3

0

4

wavelengths [nm]

rela

tive

spec

tral p

ower

di

strib

utio

n

F7

A

D65

Figure 1.1: Spectral power distribution of the Just Normlicht mini 5000 light table emulatingan F7 light source (solid lines) and the emulations of the A (dashed lines) and D65 (dottedlines) illuminants of a SpectroEye Xrite spectrophotometer.

154907), the daylight fluorescent red colorant (m f superposed with y f ) and the daylight

fluorescent green colorant (cyan superposed with y f ) printed with a Canon Pro 9500 inkjet

printer on a paper containing optical brighteners (Canon MP-101) under the considered

illuminants as well as the reflectances of the corresponding classical non-fluorescent colorants.

In Figure 1.2a we observe in the m f spectral reflectance two fluorescent peaks. The first peak

is located between 420 and 450 nm. Since the D65 illuminant has the highest energy in the UV

range (see Figure 1.1, dotted lines), it is responsible for the strongest peak at a reflectance factor

1.15. The second peak is located between 590 and 610 nm. Since the D65 and F7 illuminants

have more energy in the second part of the fluorescent excitation range (500-560 nm) than the

A illuminant, they produce the largest peak. Regarding the daylight fluorescent yellow ink y f

(Figure 1.2b), since the F7 illuminant has the highest energy in the excitation range (Figure 1.1,

solid lines), the strongest peak is observed under this illuminant with a reflectance factor of

1.62 at 520 nm. We observe similar behaviors with the daylight fluorescent green (Figure 1.2d)

and the daylight fluorescent red (Figure 1.2c) colorants. Note that the peak of the fluorescent

red at 590 nm is higher than the corresponding peak of the fluorescent magenta, due to the

additional energy absorbed due to the peak of the daylight fluorescent yellow, i.e. at 530 nm.

Finally, we observe that all daylight fluorescent colorants are more saturated and brighter

than the corresponding classical colorants. For instance, the daylight fluorescent green col-

orant (Figure 1.2d) has a narrow fluorescent peak between 500 and 540 nm with a maximal

reflectance factor of 1.15 at 530 nm and almost no reflectance at the other wavelengths. In

comparison with the classical green colorant which has a maximal reflectance factor of 0.51,

the daylight fluorescent green colorant is much more saturated and brighter.

Ink manufacturers have also developed daylight fluorescent inks for offset printers. Test

4


0

0.5

1.5

1

400 500 600 700

Ref

lect

ance

fact

ors

wavelengths [nm](a)

mf

m

0

0.5

1.5

1

400 500 600 700

Ref

lect

ance

fact

ors

wavelengths [nm](b)

yf

y

0

0.5

1.5

1

400 500 600 700

mf yf

Ref

lect

ance

fact

ors

wavelengths [nm](c)

my

0

0.5

1.5

1

400 500 600 700wavelengths [nm]

(d)

Ref

lect

ance

fact

ors

c yf

cy

F7D65AClassical inks

paperwhite (D65)

Figure 1.2: Reflectance factors of (a) the daylight fluo magenta m f colorant, (b) the daylightfluo yellow y f colorant, (c) the daylight fluo red colorant (m f superposed with y f ) and (d) thedaylight fluo green colorant (cyan superposed with y f ) under the F7 (solid lines), D65 (dottedlines) and A (dashed lines) illuminants, together with the classical colorant reflectances underthe D65 illuminant (pointed lines).

have been conducted for demonstrating the feasibility of combining classical offset inks with

daylight fluorescent offset inks. We combined classical cmyk inks with two daylight fluorescent

inks, the daylight fluorescent magenta ink (AMRA AG, ink ref. 750 - 5 - 806) and the daylight

fluorescent yellow ink (AMRA AG, ink ref. 750 - 0 - 803). The print has been carried out on an

Heidelberg 6 ink offset printer at printing company Jean Genoud SA in Lausanne, Switzerland.

With this print setup, the classical inks have to be printed before the daylight fluorescent inks.

Otherwise, due to the surface characteristics of the daylight fluorescent inks, the classical inks

do not adhere to the print.

These offset daylight fluorescent inks differ from the inkjet daylight fluorescent inks. Compared

to the inkjet m f colorant (Figure 1.3a, pointed line), the offset m f colorant (Figure 1.3a, solid

line) has not a strong fluorescent emission peak between 420 and 450 nm . The offset m f

ink seems to more absorb the paper fluorescent emission (Figure 1.3a, dashed line) at these

wavelengths. We also observe that the second emission peak located near 600 nm is slightly

higher for the offset daylight fluorescent magenta ink with a reflectance factor of 1.4. The

offset y f colorant is less saturated than the inkjet y f colorant (Figure 1.3b) with two smaller

5


fluorescent emission peaks of 0.92 and 1.18 at respectively 520 and 590 nm. When superposing

the m f ink with the y f ink (Figure 1.3c), we obtain for the offset print a strongly saturated red

colorant with a fluorescent emission peak at a reflectance factor of 1.39 at 620 nm and for the

inkjet print a strong orange colorant with two peaks at respectively 510 and 580 nm.

0

0.5

1.5

1

400 500 600 700

Ref

lect

ance

fact

ors

wavelengths [nm](a)

mf

0

0.5

1.5

1

400 500 600 700

Ref

lect

ance

fact

ors

wavelengths [nm](b)

yf

0

0.5

1.5

1

400 500 600 700

mf yf

Ref

lect

ance

fact

ors

wavelengths [nm](c)

offset paper white

simulated offset yf

inkjet mf

offset mf

inkjet yf

offset yf

inkjet mf yf

offset mf yf

Figure 1.3: Reflectance factors of (a) the daylight fluo m f colorant, (b) the daylight fluo yellowy f colorant, (c) the daylight fluo red colorant (m f superposed with y f ) measured on an offsetprint (solid lines) and on an inkjet print (pointed lines) under the D65 illuminant. The dashedlines in (b) represent the simulated offset y f colorant obtained with the inkjet y f and m f inks.

We can demonstrate that the offset y f colorant has been created by mixing at different con-

centrations the same fluorescent compounds that are used in the y f ink and in the m f inkjet

ink. We calibrate a spectral prediction model with the inkjet y f and m f colorant spectral

reflectances (see Chapter 2). We then invert the spectral prediction model in order to get

the y f and m f ink surface coverages that approximate the spectral reflectance of the offset

y f colorant. Spectral prediction model inversion is performed with a gradient descent by

minimizing the root mean square difference (RMS) between the predicted and measured

spectral reflectance of the offset y f colorant. Figure 1.3b (dashed line) shows the simulated

y f offset colorant spectral reflectance obtained by printing with a Canon Pro 9500 inkjet

printer an halftone at 100 % y f and 36% m f surface coverages. The simulated and measured

spectral reflectance of the y f offset colorant are nearly identical with a RMS difference of 0.056,

showing that the offset y f ink uses the same fluorescent compounds found in the inkjet m f

6


and y f inks. The small differences are possibly due to the different properties of the inkjet and

offset inks, such as the viscosity, the opacity, the density and the different papers used in the

two different prints.

When putting daylight fluorescent colorants under daylight illumination fading effects may

come up (Connors-Rowe et al. 2005), yielding a change of their color appearance. The fading

effect expressing the change in color is generally provided by the fluorescent pigment and/or

ink manufacturer. This effect is calculated for a specific exposure such a daylight or a fluo-

rescent lamp exposure. It can be calculated with an accelerated procedure by exposing the

sample for a short period of time under a Xenon lamp filtered for simulating the exposure.

Some testing procedures expose the sample during a long period of time and the change in

color appearance is reported (AST 2011). We perform our own experiment in order to evaluate

the fading of these fluorescent colorants. For this purpose, we let both an offset and inkjet

printed sample behind a window at a location where direct sunlight cannot illuminates di-

rectly the samples and we measure at respective intervals during one month the total spectral

reflectances of the y f and m f colorants. Figure 1.4 shows the decrease of the fluorescent emis-

sion peaks of both the inkjet and offset y f and m f colorants. In case of the offset colorants, we

do not see much difference between the original measures (blue solid lines) and the measures

performed after four weeks (blue dashed lines), with small ∆E94 differences of 1.13 for the

m f colorant and of 0.9 for the y f colorant (Table 1.1). In case of the y f inkjet colorant, it

already shows after one week a significant decrease of its fluorescent emission peak from a

reflectance factor of 1.39 to a reflectance factor of 1.20 at 520nm, yielding a strong change in

color appearance with a ∆E94 difference of 3.93 (Table 1.1). After four weeks, we observe a

huge fading effect for the inkjet y f colorant with a ∆E94 difference of 9.66. The offset m f and

y f colorants have a good lightfastness, while the lightfastness of the inkjet m f colorant is poor

and the one of the inkjet y f colorant is very poor.

Fading of the daylight fluorescent inks can be limited by printing thick layers and/or by

coating the printed layers with a UV-absorbing coating (Streitel 2009). Ink manufacturers are

developing fading resistant daylight fluorescent pigments.

Table 1.1: ∆E94 color difference between original offset and inkjet m f and y f measures andafter specific period of time under daylight behind a window glass.

∆E94

1 week 2 weeks 3 weeks 4 weeks

offset m f colorant 0.54 0.69 0.89 1.13offset y f colorant 0.51 0.59 0.74 0.9inkjet m f colorant 0.80 1.78 3.17 4.64inkjet y f colorant 3.93 6.12 8.16 9.66

7


0

0.5

1.5

1

400 500 600 700

Ref

lect

ance

fact

ors

wavelengths [nm](a)

mf

0

0.5

1.5

1

400 500 600 700

Ref

lect

ance

fact

ors

wavelengths [nm](b)

yf2-5

inkjet

offset

inkjet

offset

Figure 1.4: Reflectance factors of (a) the daylight fluorescent m f colorant and (b) the daylightfluorescent y f colorant under the D65 illuminant. Black lines show the inkjet colorants andblue lines show the offset colorant, with original measured spectral reflectances as solid linesand after four weeks as dashed lines.

1.4 Dissertation outline

Chapter 2 starts by establishing a spectral prediction model dedicated for predicting the

spectral reflectances of halftones combining classical and daylight fluorescent inks. The

accuracy of this model is compared to the accuracy of classical spectral prediction models

such as the cellular Yule-Nielsen model (CYNSN) and the ink spreading enhanced Yule-Nielsen

model (IS-YNSN). Accuracy is computed as ∆E94 prediction error for uniformly distributed

halftones printed with inkjet, electrophotographic and offset printers.

In Chapter 3 we then compare for an inkjet printer the gamuts of a classical cmyk ink print

setup and of a fluorescent 6 ink print setup comprising the classical cmyk inks and the daylight

fluorescent magenta and yellow inks. The gamuts are compared under the A, D65 and F7

illuminants as CIELAB constant lightness planes and by calculating the additional volume

of the sRGB display gamut covered by the 6 ink fluorescent print setup. We also compare as

CIELAB constant lightness planes the inkjet and offset fluorescent 6 ink gamuts.

In Chapter 4 we describe gamut expansion and reduction algorithms used to map input

sRGB images to destination fluorescent printable images. We describe the advantages and

the disadvantages of different possible projections. In addition, we show how to control

gamut mapping expansion and reduction from the display sRGB gamut to the printable

fluorescent ink gamut. The mapping is controlled by user-defined parameters such as the type

of function used to map input lightnesses to destination lightnesses, a non-linear mapping

chroma reinforcement factor controlling how fast the chroma expansion is applied, a gamut

expansion limitation factor that limits the maximal effective gamut expansion, the focal

points giving the direction of mapping lines and means of creating smooth chroma transitions

between expanded and non-expanded image parts. We also show how to display preview the

destination gamut expanded image.

8

1.5. Contributions

In Chapter 5 we show results of the gamut expansion and reduction of sRGB images by

comparing photographs of both offset and inkjet printed images either with classical cmyk

inks only or with the fluorescent 6 ink print setup.

In Chapter 6 we show how to use daylight fluorescent magenta and yellow inks in order to

hide under a specific illuminant security patterns or variable intensity images within printed

images and to reveal them under an illuminant different from the reference illuminant. We

also show how to hide the security patterns under various common natural and artificial

illuminations.

We draw the conclusions in Chapter 7.

1.5 Contributions

The list of contributions of this thesis are described as follow:

- We propose an extension of the cellular Yule-Nielsen spectral Neugebauer model by

accounting for ink spreading of each ink within each subdomain. Ink spreading is

characterized by jointly fitting within each subdomain the ink spreading functions on a

single halftone patch located at the center of the considered subdomain. In case of three

ink halftones, compared to the cellular Yule-Nielsen model, this extension requires only

8 additional spectral reflectance measurements.

- We show that this new ink spreading extension of the cellular Yule-Nielsen model

is remarkable accurate for predicting the spectral reflectances of classical halftones

printed with inkjet, electrophotographic and offset printers and for predicting combined

classical ink and daylight fluorescent ink hafltones printed with inkjet and offset printers.

- We show that ink spreading can be characterized with tri-stimulus sensor responses

instead of using full spectral measurements without reduction of prediction accuracy.

- We propose a complete solution for printing with a 6 ink print setup combining classi-

cal and daylight fluorescent inks. The solution comprises a spectral prediction model

dedicated for the accurate spectral prediction of halftones combining classical and day-

light fluorescent inks, a gamut mapping from the sRGB display to the partly fluorescent

printer gamut, and a generation of the 6 ink color separation layers.

- We propose a user-driven gamut expansion method for mapping sRGB images to print-

able fluorescent ink gamut images. The expansion ensures continuity of colors of the

input image and the gamut expansion parameters allow controlling the chroma expan-

sion of image parts as well as creating smooth chroma transitions between expanded

and non-expanded image parts.

- We propose a solution for previewing on a standard sRGB display destination gamut

expanded images having colors beyond the sRGB display gamut.

9


- We show how to hide under a specific illumination security patterns within printed im-

ages with the 6 ink print setup combining classical and daylight fluorescent inks. These

patterns are revealed under an illumination different from the reference illumination

used to hide the security patterns.

- We show how to hide security variable intensity images within printed images. These

security images are revealed under an illumination different from the reference illumi-

nation.

- We show how to hide security patterns under various common natural and artificial

illuminations. These security patterns are revealed under an illuminant having energy

only in the excitation wavelengths of the daylight fluorescent inks.

1.6 Review of the prior art

A printer is characterized by the relationship between the printer’s input in terms of nomi-

nal surface coverages of the inks and the resulting output colors. This relationship is often

obtained by printing hundreds of color halftones at different combinations of ink surface

coverages. Each sample is measured by a spectrophotometer under the reference illumination

and converted to a color value. One may then interpolate between these color values to create

the mapping between desired color and surface coverages of the inks, see (Bala 2003a).

Another approach consists in modeling the interaction of the light and the print according

to a spectral prediction model. At the present time, among the existing spectral reflection

prediction models, mainly the well-known Yule-Nielsen modified spectral Neugebauer model

(YNSN) (Yule and Nielsen 1951), (Viggiano 1990) is used for predicting reflection spectra (Bala

1999), (Wyble and Berns 2000), (Ogasahara 2004). Due to the printing process, the deposited

ink dot surface coverage is generally larger than the nominal surface coverage, yielding a

"physical" dot gain responsible for the ink spreading phenomenon (Bala 1999). In order to

make accurate spectral prediction, the YNSN model needs to take into account ink spreading

of the ink halftone dots. Most of the time, ink spreading is accounted for each ink separately

through a single tone reproduction curve (Pobboravsky and Pearson 1972). However, it has

been shown by Hersch et al. that dot gain also depends on the ink superposition conditions.

They proposed an ink spreading extension of the Yule-Nielsen model, the ink spreading

enhanced Yule-Nielsen modified spectral Neugebauer model (IS-YNSN) (Hersch and Crété

2005) by establishing for each ink in the different superposition conditions, i.e. alone on paper,

in superposition with one ink, in superposition with two or more inks, ink spreading curves

mapping nominal to effective surface coverages.

In order to provide a higher prediction accuracy, Heuberger et al. (1992) proposed the Cellular

Neugebauer Model. Bala (1999) has shown that the cellular subdivision is also applicable to

the Yule-Nielsen spectral Neugebauer model (CYNSN). In prior work, the cellular Yule-Nielsen

model was further improved along the following lines:

10

1.6. Review of the prior art

(a) Optimization of the Neugebauer primary reflectances according to the color halftone

patches forming the learning set (Bala 1999).

(b) Octtree like hierarchical subdivision of the surface coverage cube and subcubes until

the desired prediction accuracy is reached (Agar and Allebach 1998).

(c) Introducing for each ink a single function relying on single ink halftone ramps mapping

nominal to effective coverages (Bala 1999) (Chen et al. 2004).

(d) Optimization of the positions for the non-uniform cellular subdivisions of the surface

coverage unit cube (Chen et al. 2004).

In the past, there was an attempt to print with classical and daylight fluorescent inks. Guyler

(2001) compared the gamuts of classical and combined classical and daylight fluorescent inks

for offset prints by relying on Neugebauer primaries and on printed color patch measurements.

However, no attempt to create a full color management framework combining classical and

daylight fluorescent inks was proposed.

Printing with combined classical and daylight fluorescent custom inks faces similar problems

as printing with custom inks. There is a need to select a specific subset of inks from many

possible subsets and to map the input gamut into the gamut achievable with the multi-ink

hafltones. Stollnitz, Ostromoukhov, and Salesin (1998) modeled the gamut of printable custom

colorants by a modified Neugebauer model accounting for trapping, dot gain and multiple

internal reflectances. With this model, they optimized the selection of custom inks in order to

obtain a given color. Tzeng and Berns (2000) used cyan, magenta, yellow, black, orange and

green inks and developed an algorithm for selecting a subset of 4 inks among the 6 inks to

reproduce a given reflection spectrum as accurately as possible, by minimizing metamerism.

For generating gamut expanded images there is a need of mapping an input gamut, e.g. the dis-

play sRGB gamut, to the fluorescent printable destination gamut. Traditional gamut mapping

techniques are presented by Morovic and Luo (2001). Kang et al. (2003) performed a user study

which in addition to gamut compression also dealt with users performing interactive gamut

expansion from print gamut to display gamut. For most colors, besides memory colors, the

users tried to extend the chroma of the images. Hirokawa et al. (2007) showed that, compared

with the linear expansion, non-linear chroma expansion of sRGB images displayed on a wide

gamut Adobe RGB monitor was preferred by users.

Another application of fluorescence is the authentication of security documents (Van Renessse

2005). Bala et al. (2007) used the fluorescence of paper incorporating fluorescent brighteners

in order to create images embedding security information that is invisible under normal

daylight and revealed under UV illumination. Security features relying on single invisible

fluorescent inks are widely used in passports, bank notes and credit cards (Van Renessse 2005).

The hidden patterns are generally printed with a single invisible fluorescent ink, for example

the yellow "VISA" appearing on Visa credit cards under a UV light source. Hersch et al. (2007)

11


proposed to enhance the security provided by invisible fluorescent inks by creating full color

images viewable under UV light with three inks having their fluorescent emission in different

parts of the visible wavelength range.

Coyle and Smith (2004) propose to create fluorescent color images with red, green and blue

emitting fluorescent inks, which are invisible under daylight. Narita and Eto (2002) teach

how to form an image with color gradations using fluorescent red, green and blue colorants,

colorless under normal daylight and emitting fluorescence under UV illumination. Brehm

and Erbar (2001) also describe an additive fluorescent ink mixing process capable of creating

continuous tone halftone images. Jones II et al. (2002) describe a method of marking products

by tags formed by luminescent inks having specific emission wavelength ranges and specific

decay times. Auslander and Cordery (2003) propose a print head system with a first ink having

a first color under normal daylight and a second fluorescent ink having the same color as the

first ink under normal daylight but discernible from the first ink when subjected to fluorescent-

exciting radiation. This second ink is visible only under an exciting radiation enables creating

covert markings. Auslander et al. (2002) teach a method for printing a security marking with

an ink absorbing light under daylight (dark patterns) and emitting light under an excitation

illumination. This security marking is viewed both under daylight and by fluorescence under

fluorescent excitation illumination.

However, these security features rely on invisible fluorescent ink which do not absorb in the

visible wavelength range. In contrast to these methods, we propose to hide security patterns

by printing image parts with combinations of classical and daylight fluorescent inks while the

rest of the image is printed with combinations of classical inks only. In addition, we apply a

metameric color match between the image part printed with and without daylight fluorescent

inks.

12

2 Prediction model for classical anddaylight fluorescent inks

2.1 Introduction

The total reflectances shown in Chapter 1 Section 1.3 indicate that new saturated and bright

colorants can be obtained either by combining classical and daylight fluorescent inks or by

combining two daylight fluorescent inks. The current chapter describes how to accurately

predict the spectral reflectances of halftones comprising daylight fluorescent inks with an

ink spreading extension of the cellular Yule-Nielsen spectral prediction model (IS-CYNSN).

Accurate spectral predictions are needed for establishing a mapping between CIELAB colors

and corresponding ink dot surface coverages and for establishing the exact printer gamut.

We show that although this new ink spreading extension of the cellular Yule-Nielsen model

is calibrated with a few halftone patch reflectances, i.e. only 35 halftone patch reflectances

for predicting 3 ink halftones, this model is remarkably accurate for predicting spectral re-

flectances of both halftones comprising classical inks only printed with inkjet, offset and

electrophotographic printers (Section 2.5) and halftones combining classical and daylight

fluorescent inks printed with inkjet and offset printers (Section 2.6).

2.2 Limitation of the Yule-Nielsen model for predicting fluorescent

ink halftones

Within halftones comprising classical and daylight fluorescent inks, light propagates from

one fluorescent colorant to a non-fluorescent colorant and vice versa. The emitted part of

fluorescent ink halftone element depends on the absorption of its neighbouring halftone

elements. In this situation, the Yule-Nielsen modified Neugebauer model (Yule and Nielsen

1951) (Viggiano 1990) is not accurate for predicting the total spectral reflectances of halftones

comprising daylight fluorescent inks. To illustrate the low prediction accuracy of the Yule-

Nielsen model, we printed all combinations of the cm f y f inks by varying the nominal ink

surface coverages by steps of 25%, yielding 53 = 125 halftones. Halftones were printed with an

inkjet Epson P50 printer and with an offset printer. We calibrate an ink spreading enhanced

13

Chapter 2. Prediction model for classical and daylight fluorescent inks

Table 2.1: Prediction accuracies of the IS-YNSN model under the D65 illuminant for 125 cm f y f

test samples printed with an inkjet Epson P50 printer and an offset printer.

Test set ∆E94

Avg 95% Max

inkjet cm f y f 2.47 5.19 6.60offset cm f y f 3.37 8.88 12.63

Yule-Nielsen spectral Neugebauer model (IS-YNSN) (Hersch and Crété 2005) by measuring

44 uniform halftone samples. We then run a spectral prediction for all the 125 halftones. The

prediction accuracies for the offset and inkjet printed fluorescent halfltones are listed in Table

2.1. We obtain under the D65 illuminant for the inkjet and offset prints the respective CIELAB

mean ∆E94 prediction errors of 2.47 and 3.37 and respective 95% quantile prediction errors

of 3.8 and 8.88. We also observe for the offset print a huge maximal ∆E94 prediction error

of 12.63. These prediction accuracies are not sufficient in order to create precise colors by

deriving from the model the surface coverages of the inks.

2.3 The ink spreading enhanced cellular Yule-Nielsen model

As an alternative, we established an ink spreading extension of the cellular Yule-Nielsen model

(named: IS-CYNSN) (Rossier et al. 2010). The IS-CYNSN model has a finer ink surface coverage

space subdivision and therefore better accounts for the fluorescent emission. In addition,

the IS-CYNSN model accounts for the ink spreading phenomenon, yielding accurate spectral

predictions. Finally, the IS-CYNSN model needs to be calibrated with only a limited number

of spectral reflectance measurements. In the case of halftones formed by 3 inks, we only need

35 spectral reflectance measurements and in the case of halftones formed by 4 inks, we only

need 97 spectral reflectance measurements.

2.3.1 Ink spreading extension of the Cellular Yule-Nielsen model

The Yule-Nielsen modified Neugebauer spectral model (Equation 2.1) is used to predict the

spectral reflectance R(λ) of a color halftone as a weighted sum of Neugebauer primary re-

flectances Ri (λ), where ai is the area coverage of the i th primary, Ri (λ) its reflection spec-

trum and n the Yule-Nielsen value accounting for the lateral propagation of light (in general,

1 < n < 100).

R(λ) =(∑

iai ·Ri (λ)1/n

)n

(2.1)

14

2.3. The ink spreading enhanced cellular Yule-Nielsen model

Hereinafter, we focus on the spectral reflectance prediction models for three inks. However,

the models can be extended to 4 inks (Bugnon et al. 2008). With three inks, we have 23 = 8

primaries corresponding to all combinations of 0% and 100% ink surface coverages. Assuming

independently printed cyan, magenta and yellow inks, the area coverages ai of the primaries

white (ai ), cyan (ac ), magenta (am), yellow (ay ), red (ar ) (superposition of cyan and yellow),

blue (ab) (superposition of cyan and magenta) and black (ak ) (superposition of cyan, magenta

and yellow) are calculated according to the Demichel’s equations (Wyble and Berns 2000)

expressed by Eqs. 2.2

aw = (1− c)·(1−m)·(1− y) ac = c ·(1−m)·(1− y)

am = (1− c)· m ·(1− y) ay = (1− c)·(1−m)· y

ar = (1− c)· m · y ag = c ·(1−m)· y

ab = c · m ·(1− y) ak = c · m · y

(2.2)

where c, m, y represent respectively the cyan, magenta and yellow ink surface coverages.

These 8 area coverages are identical to the 8 coefficients used for tri-linear interpolation

between known cube vertex values.

(0,0,0) (0.5,0,0)

(0.5,0.5,0)(0,0.5,0)

(0,0,0.5)

(0.5,0.5,0.5)(0,0.5,0.5)

(0.5,0,0.5)

cyan

yellow

magen

ta

Figure 2.1: Illustration of the cellular Yule-Nielsen model where the illustrated cube representsone of the 8 subdomains produced by all combinations of 0%, 50% and 100% surface coveragesof the three inks. At the vertices of the cube, subdomain primary reflectances Rc,m,y (λ) havebeen measured.

Thanks to the cellular Yule-Nielsen extension of the Neugebauer model (Bala 1999), we im-

prove the prediction accuracy by dividing the CMY ink surface coverage space into 8 subdo-

mains. As Neugebauer primaries, we not only consider reflectances of printed haltones at

0% and 100% surface coverages, but also printed halftones (called subdomain primaries) at

all combinations of 0%, 50% and 100% surface coverages (23 = 27 combinations). Figure 2.1

illustrates a subdomain where the cyan, magenta and yellow ink surface coverages vary from 0

to 0.5. For ink surface coverages within that subdomain, we first normalize the subdomain

coverages. With c, m, y ink surface coverages of cyan, magenta and yellow between 0 and 0.5,

15


the normalized coverages c ′, m′ and y ′ are

c ′ = c

0.5m′ = m

0.5y ′ = y

0.5(2.3)

The areas of subdomain primaries are calculated from the normalized coverages c ′, m′ and y ′

with coefficients expressed by the Demichel’s equations (Eqs. 2.2). The spectral prediction is

carried out by tri-linear interpolation, i.e. by weighting the subdomain primary reflectances

with the corresponding areas of subdomain primaries according to the Yule-Nielsen equation

(Equation 2.1).

More precisely, for an arbitrary cellular subdivision and with cyan, magenta and yellow ink

surface coverages c, m, y within the subdomain delimited by c ∈ [cl ,ch], m ∈ [ml ,mh] and

y ∈ [yl , yh], the normalized c ′,m′ and y ′ ink coverages are

c ′ = c − cl

ch − clm′ = m −ml

mh −mly ′ = y − yl

yh − yl(2.4)

The predicted reflectance R(λ) of a halftone of surface coverages c ∈ [cl ,ch], m ∈ [ml ,mh],

y ∈ [yl , yh] is obtained by tri-linear interpolation of cube vertex reflectances

R(λ) = ((1− c ′) · (1−m′) · (1− y ′) ·Rcl ,ml ,yl (λ)1/n

+ c ′ · (1−m′) · (1− y ′) ·Rch,ml ,yl (λ)1/n + (1− c ′) ·m′ · (1− y ′) ·Rcl ,mh,yl (λ)1/n

+ (1− c ′) · (1−m′) · y ′ ·Rcl ,ml ,yh(λ)1/n + (1− c ′) ·m′ · y ′ ·Rcl ,mh,yh(λ)1/n

+ c ′ · (1−m′) · y ′ ·Rch,ml ,yh(λ)1/n + c ′ ·m′ · (1− y ′) ·Rch,mh,yl (λ)1/n

+ c ′ ·m′ · y ′ ·Rch,mh,yh(λ)1/n)n

(2.5)

where Rc,m,y (λ) represents the measured spectral reflectance at surface coverages (c,m, y) of

the cyan, magenta and yellow inks. The prediction accuracy of the cellular Yule-Nielsen model

can be improved by a finer subdivision or by multiple levels of subdivisions. For instance,

we can increase the number of subdomains by choosing for the subdomain primaries the

combinations of 0%, 25%, 50%, 75% and 100% nominal ink surface coverages. In this case, the

number of required subdomain primary spectral measurements increases significantly (in the

present case: 53 = 125 spectral measurements).

In order to improve the prediction accuracy, as an alternative to the increase of subdomains,

we propose an ink spreading extension of the cellular Yule-Nielsen model where ink spreading

is accounted for within each subdomain. We create within each subdomain ink spreading

curves expressing the ink spreading behavior of the ink hafltone dots. Since the dot gain of

16

2.3. The ink spreading enhanced cellular Yule-Nielsen model

one ink within a subdomain does not depend strongly on the other ink surface coverages,

we only consider the ink spreading of each ink for a single ink superposition condition. This

yields, for each ink i within each subdomain j , an ink spreading curve fi , j (u′i , j ) mapping the

normalized ink coverage u′i , j to a normalized effective ink coverage u′

i , j ,eff . The ink spreading

curves are obtained by printing halftones in one ink superposition condition, i.e. with one ink

at a nominal surface coverage corresponding to the mid-range of the considered subdomain

and the other inks at their lower bounds. For instance, the ink spreading curve for the cyan ink

(i = c) within the subdomain j delimited by its low (l ) and high (h) bounds ui=c, j ∈ [c j l ,c j h],

ui=m, j ∈ [m j l ,m j h], and ui=y, j ∈ [y j l , y j h] is established by printing a halftone at cyan mid-

range, magenta low bound and yellow low bound ink nominal surface coverages, i.e. a halftone

at cyan ui=c, j = (c j l+c j h)/2 at magenta ui=m, j = m j l and at yellow ui=y, j = y j l . Then, we fit the

mid-range cyan normalized effective surface coverage qi=c, j of subdomain node reflectances

by minimizing the sum of square differences between measured halftone reflection spectrum(Ri=c, j (λ)

)and the corresponding predicted reflectance spectrum

(Ri=c, j (λ)

)

Ri=c, j (λ) = R(c j l+c j h )/2,m j l ,y j l (λ)

Ri=c, j (λ) = (qi=c, j ·Rc j h ,m j l ,y j l (λ)1/n + (1−qi=c, j ) ·Rc j l ,m j l ,y j l (λ)1/n

)1/n

qi=c, j = argmin∑

k[Ri=c, j (λk )− Ri=c, j (λk )

]2(2.6)

The minimization can be carried out with a computer executable procedure implementing

Powells’ function minimization (Press et al. 1998). The two other qi=m, j and qi=y, j ink normal-

ized effective surface coverages of subdomain node reflectances are obtained by replacing the

cyan mid-range by the corresponding ui , j mid-range, the cyan higher bound c j h by the ui , j

higher bound and keeping in Equation 2.6 the other ink coverages at their lower bound. The

fitted ink normalized effective surface coverage qi , j indicates the amount of ink spreading of

ink i within the subdomain j . The ink spreading curves u′i , j ,eff = fi , j (u′

i , j ) within the subdo-

main j are obtained by quadratic interpolation between the points (0,0), (0.5, qi , j ) and (1,1),

with u′i , j ,eff = (2−4 ·qi , j ) ·u′

i , j2 + (4 ·qi , j −1) ·u′

i , j .

Computing the normalized effective surface coverages qi=c, j , qi=m, j and qi=y, j with Equation

2.6 requires for each subdomain j three spectral reflectance measurements, i.e. we have to

use for each ink i an halftone at nominal surface coverage of the ink i corresponding to the

mid-range of the considered subdomain and the two other inks at their lower subdomain

bound. As an alternative, in order to decrease the number of reflectance measurements to

one per subdomain, we propose to jointly fit the normalized effective surface coverages on a

17


single halftone located at the center of the considered subdomain j (Equation 2.7)

Ri=center, j (λ) = R(c j l+c j h )/2,(m j l+m j h )/2),(y j l+y j h )/2(λ)

Ri=center, j (λ) = ((1−qi=c, j )·(1−qi=m, j )·(1−qi=y, j )·Rc j l ,m j l ,y j l (λ)1/n

+ qi=c, j ·(1−qi=m, j )·(1−qi=y, j )·Rc j h ,m j l ,y j l (λ)1/n

+ (1−qi=c, j ) · qi=m, j ·(1−qi=y, j )·Rc j l ,m j h ,y j l (λ)1/n

+ (1−qi=c, j ) ·(1−qi=m, j )· qi=y, j ·Rc j l ,m j l ,y j h (λ)1/n

+ (1−qi=c, j ) · qi=m, j · qi=y, j ·Rc j l ,m j h ,y j h (λ)1/n

+ qi=c, j ·(1−qi=m, j )· qi=y, j ·Rc j h ,m j l ,y j h (λ)1/n

+ qi=c, j · qi=m, j ·(1−qi=y, j )·Rc j h ,m j h ,y j l (λ)1/n

+ qi=c, j · qi=m, j · qi=y, j ·Rc j h ,m j h ,y j h (λ)1/n)n{

qi=c, j , qi=m, j , qi=y, j} = argmin

∑k[Ri=center, j (λk )− Ri=center, j (λk )

]2

(2.7)

Figure 2.2 illustrates a cyan dot gain curve for a cmy laser print, where the normalized dot gain

is defined as di , j (u′i , j ) = fi , j (u′

i , j )−u′i , j , within the subdomain j = 1 delimited by c ∈ [0,0.5],

m ∈ [0,0.5] and y ∈ [0,0.5]. The computed cyan normalized effective surface coverage qi=c,1

of subdomain node reflectances for an optimal n-value = 14 calculated with Equation 2.6 is

equal to 0.61. It represents a normalized dot gain of 0.11 in the range [0,1] and therefore a real

dot gain of 0.055 in the range [0,0.5]. The cellular Yule-Nielsen model prediction error for the

considered uniform simple halftone without taking into account the dot gain is ∆E94 = 3.60.

Introducing the dot gain obtained by the fitted cyan normalized effective surface coverage

qi=c, j of subdomain node reflectances decreases for this halftone the prediction error to

∆E94 = 0.22. Since for the present print configuration, the ink dot gain within each subdomain

j is at least 0.1, accounting for ink spreading considerably increases the spectral prediciton

accuracy.

0 0.5 1

0.020.040.060.080.10

dot gain di=c,j=1 qi=c,j=1-0.5

normalized nominal surface coverages

Figure 2.2: Cyan dot gain curve corresponding to the cyan ink spreading curve within thesubdomain c, m, y ∈ [0,0.5], for a cmy laser print (Brother 4000-HL) at a screen frequency of120lpi and using an optimal n-value of 14.

When computing the normalized effective surface coverages qi , j of all subdomains j according

to Equation 2.7 instead of Equation 2.6, the coefficients are similar, i.e. the dot gains do not

18

2.4. Characterizing ink spreading with sensor responses

deviate by more than 10%. We therefore obtain the same prediction accuracy improvements

by jointly fitting the normalized effective surface coverages on a single center subdomain

halftone relfectance as when fitting them on the three spectral reflectance measurements

required by Equation 2.6.

Note that the optimal n-value is found by predicting for successive n-values with the full

model all mid-range reflectances. The n-value yielding the minimal average prediction error

is kept as the optimal n-value for the considered setup of printer, inks and paper. Since we

already measured the spectral reflectances of the mid-range for computing the normalized

effective surface coverages at model calibration, the n-value can be fitted without measuring

additional haftone patch reflectances.

The cellular Yule-Nielsen model accounting for ink spreading (ISseparately-CYNSN and ISsingle-

CYNSN) is illustrated in Figure 2.3. At model calibration, the subdomain ink spreading curves

fi , j (u′i , j ) are established either by separately fitting the normalized effective surface coverages

with Equation 2.6 (ISseparately-CYNSN) or by jointly fitting the normalized effective surface

coverages with Equation 2.7 (ISsingle-CYNSN). At run time, nominal ink surface coverages of

the considered halftone are normalized according to Equation 2.4, the normalized effective

ink surface coverages are deduced by making use of the corresponding ink spreading curves,

the normalized effective areas of the subdomain primary reflectances are calculated according

to Equation 2.2 and the halftone reflection spectrum is predicted according to Equation 2.1.

c m y

fi=m,j (mj’)

Normalization within sub-domain jcj’ mj’ yj’

cj’,eff mj’,eff yj’,eff

fi=c,j (cj’) fi=y,j (yj’)

Calculation of eff. primary coveragesajw’ ajc’ ajm’ ajy’ ajr’ ajg’ ajb’ ajk’

Spectral prediction

Predicted reflection spectrum

sub-domain primaryreflectances

Figure 2.3: Cellular Yule-Nielsen model accounting for ink spreading.

2.4 Characterizing ink spreading with sensor responses

The three ink non-cellular ink spreading Yule-Nielsen modified spectral Neugebauer model

(IS-YNSN) is calibrated with 8 Neugebauer primaries (Hersch and Crété 2005). In addition,

in order to account for ink spreading in all ink superposition conditions, 12 ink spreading

curves are established mapping nominal surface coverages to effective surface coverages.

Establishing the ink spreading curves requires measuring hafltones of each ink in its 4 possible

19


ink superposition conditions, i.e. alone on paper, superposed with one solid ink, superposed

with the other solid ink and superposed with the two others solid inks. In this setup, the model

requires 8+12 = 20 spectral measurements. In the case of the ink spreading cellular Yule-

Nielsen models, with one level of subdivision we obtain 27 subdomain primaries (subdomain

primaries are considered by taking all combinations of 0%, 50% and 100% surface coverages

yielding 23 = 27 combinations). Ink spreading is modeled by establishing 3 ink spreading

curves within each of the 8 subdomains (Section 2.3.1). Thus, in case of separately fitted

coefficients, the ISseparately-CYNSN model requires 27+8 ·3 = 51 measurements and in case

of jointly fitted normalized effective surface coverages, the ISsingle-CYNSN model requires

27+8 ·1 = 35 measurements.

However, since establishing the ink spreading curves requires fitting only one scalar variable

(ISseparately-CYNSN) or three scalars (ISsingle-CYNSN) at a time, it is possible to use for exam-

ple Red, Green and Blue sensor responses for the fitting process (Garg et al. 2008). The ink

spreading characterization of the IS-YNSN, the ISseparately-CYNSN and the ISsingle-CYNSN is

performed by minimizing the sum of square differences between predicted and measured sen-

sor response values. Fitting with sensors considerably reduces the number of required spectral

measurements. Only the spectral measurements of Neugebauer primaries are necessary, 8 for

the IS-YNSN model and 27 for a single level subdivision ISseparately-CYNSN and ISsingle-CYNSN

models. In order to demonstrate the feasibility of using sensor responses instead of spectral

measurements, we simulate the RGB sensor devices by the DIN-16536-2 standard RGB sensi-

tivities for densitometric measurements (DIN 1995). Samples are illuminated with a standard

CIE D65 illuminant. Reflected light generates the sensor responses Ci

Ci =∑k

Si (λk ) ·R(λk ) · I (λk )/(

I (λk ) ·S(λk ))

(2.8)

where Ci represents the i th sensor response values, Si the spectral sensitivity of the i th sen-

sor, I (λ) the illuminant and the R(λ) the spectral reflectance (Bala 2003b). We can fit the

normalized effective surface coverages qi , j of subdomain node reflectances by replacing in

Equation 2.6, respectively in Equation 2.7 predicted Ri , j and measured Ri , j reflectances by

their corresponding Ci , j and Ci , j sensor responses, according to Equation 2.8.

Let us consider as example the cmy laser print. The three dot gain curves within the subdomain

shown in Figure 2.1 are similar one to another. The normalized effective surface coverages

qi=c, j=1, qi=m, j=1 and qi=y, j=1 of node reflectances are either respectively equal to 0.6050,

0.5989 and 0.5912 when fitted with the spectral reflectance metric or respectively equal to

0.6069, 0.6000 and 0.5945 when fitted with the simulated RGB sensor response metric. The

normalized effective surface coverages of subdomain node reflectances qi , j in all subdomain

j do not deviate by more than 1.5% when comparing these two metrics. Therefore, the

prediction accuracy remains the same when characterizing ink spreading for the ISseparately-

20

2.5. Prediction accuracies for classical ink halftones

CYNSN, the IS-YNSN and the ISsingle-CYNSN models with three sensor responses instead of

spectral measurements.

2.5 Prediction accuracies for classical ink halftones

We performed spectral predictions with the cellular Yule-Nielsen model, the ink spreading

enhanced cellular Yule-Nielsen models and the ink spreading enhanced Yule-Nielsen model

(Table 2.2). In order to compare the resulting prediction accuracies with prior work, we

also consider for each ink a single global ink spreading function, which is fitted with the

YNSN model at 25%, 50%, 75% nominal ink surface coverages. We also performed spectral

predictions by characterizing ink spreading with simulated RGB sensors for both the IS-CYNSN

and the IS-YNSN models (Table 2.3). The experiment were performed on an inkjet printer

(Canon Pixma Pro 9500 at 600 dpi) with standard cyan, magenta and yellow inks printed

on Canon MP-101 paper at a screen frequency of 120 lpi. In addition, test samples were

printed with a laser printer (Brother 4000-HL at 600 dpi) with standard cyan, magenta and

yellow toners on Canon MP-101 paper at a screen frequency of 120 lpi. The test samples

were printed at all combinations of nominal ink surface coverages 0, 0.25, 0.5, 0.75 and 1

(53 = 125 test patches) with classical rotated screen. Relfectances were measured with a

GretagMacBeth Color i7 spectrophotometer with geometry (d : 80) under a D65 illuminant.

In Tables 2.2 and 2.3, we give the mean prediction error in terms of ∆E94 values, the maximal

prediction error, the 95% quantile prediction error, the number of spectral primary reflectance

measurements (p) and the number of ink spreading measurements (i ). The n-value yielding

the best prediction accuracies for all models and test sets is 14.

The spectral prediction based on the proposed ink spreading extension for the cellular Yule-

Nielsen model (ISseparately-CYNSN) provides a significantly higher prediction accuracy com-

pared to the stand-alone cellular Yule-Nielsen model (CYNSN). The ∆E94 mean prediction

error decreases from 2.29 to 1.06 for a laser print (Table 2.2, Brother 4000-HL test set) and

from 0.92 to 0.54 for a classical CMY inkjet print (Table 2.2, Canon pro 9500 test set). When

printing with a laser printer (Brother 4000-HL), we observe a strong dot gain within all surface

coverage subdomains. Therefore, for this printer, there is a large difference in prediction accu-

racy between the ISseparately-CYNSN model that accounts for ink spreading and the CYNSN

model that does not account for ink spreading. In addition, considering ink spreading within

each subdomain also offers a higher prediction accuracy than when using within the CYNSN

model for each ink a single global function mapping nominal to effective surface coverages

(ISglobal-CYNSN), i.e. the ∆E94 mean prediction error decreases from 1.30 to 1.06 for the laser

print and from 0.70 to 0.54 for the inkjet print.

Introducing ink spreading within each subdomain by jointly fitting the normalized effec-

tive surface coverages provides higher prediction accuracies than by separately fitting the

normalized effective surface coverages. For instance, in case of the Canon Pro 9500 print,

with the ISseparately-CYNSN model, we obtain a ∆E94 mean prediction error of 0.54 and 95%

21


quantile prediction error of 1.49 and with the ISsingle-CYNSN model these prediction errors

decrease respectively to 0.38 and 0.99. This can be explained by the fact that the center of

each subdomain contains the most useful information in respect to the ink spreading phe-

nomenon. The ISsingle-CYNSN represents therefore an excellent tradeoff between number of

measurements and prediction accuracy. With only 35 spectral reflectance measurements we

obtain remarkable prediction accuracies. Similar test have been conducted on an offset print,

a proofing device (Kodak Approval) and other inkjet prints. In all cases, mean prediction ∆E94

error around 0.4 have been obtained. In this situation, for the rest of this thesis, we consider

ink spreading for the cellular Yule-Nielsen by jointly fitting the normalized effective surface

coverages only. The IS-CYNSN model therefore will always refer to the ISsingle-CYNSN model.

We also remarked that the ink spreading enhanced Yule-Nielsen model that accounts for ink

spreading without cellular subdivision is more accurate than the cellular Yule-Nielsen model.

This can be explained by the fact that the ink spreading behavior of multi-ink halftones is

well captured by the Yule-Nielsen spectral Neugebauer model enhanced to account for ink

spreading in all superposition conditions and that ink spreading has a strong impact on the

resulting printed color.

The prediction accuracies obtained when characterizing ink spreading with RGB sensors for

the ISseparately-CYNSN, the ISsingle-CYNSN, the ISglobal-CYNSN and the IS-YNSN models (Table

2.3) are nearly identical with the ones obtained with ink spreading characterized by spectral

measurements (Table 2.2). This shows that ink spreading characterization can be performed

by making use of RGB sensors instead of using spectral reflectance measurements. The cost of

including RGB sensors within printers is much lower compared with the cost of including a

spectrophotometer. Therefore, the ink spreading enhancement of the non-cellular as well as

the cellular Yule-Nielsen models offers the potential of characterizing printers at run time at a

moderate cost.

2.6 Prediction accuracies for fluorescent ink halftones

In order to create a 6 ink print setup using the classical cyan ink c, magenta ink m, yellow ink

y , black ink k and the two daylight fluorescent magenta m f and yellow y f inks, we define the

total printer gamut G f as the conjunction of the four cmyk, cm f y f , cm f y and cmy f ink set

sub-gamuts. In order to establish these exact sub-gamuts, we need a spectral prediction model

that is accurate for predicting colors combining the inks present in the different sub-gamut

ink sets. In order to test the prediction accuracies of the IS-CYNSN, the IS-YNSN and the

CYNSN models for the considered fluorescent 6 ink print setup, we print for each ink set all

the ink combinations by varying the ink nominal surface coverages by steps of 25%, yielding

53 = 125 halftones for the sets comprising three inks (cm f y f , cm f y and cmy f ) and 54 = 625

halftones for the set comprising 4 inks (cmyk). The halftones were printed both with an offset

printer on a HEAVEN 42 SOFTMATT coated paper and with an Epson P50 inkjet printer on

a Canon MP-101 paper. There were measured under the D65 illuminant with a SpectroEye

22

2.6. Prediction accuracies for fluorescent ink halftones

Table 2.2: Prediction accuracies for cyan, magenta, yellow test samples printed with a CanonPro 9500 inkjet printer and for cyan, magenta and yellow test samples printed with a Brother4000-HL laser printer.

Test sets # measurements ∆E94

Model p + i Avg 95% Max

Brother 4000-HLPrior art:

CYNSN 27+0 = 27 2.29 5.22 6.53ISglobal-CYNSN 27+9 = 36 1.30 3.31 3.79IS-YNSN 8+12 = 20 1.86 4.10 5.01

new:ISseparately-CYNSN 27+24 = 51 1.06 3.06 3.76ISsingle-CYNSN 27+8 = 35 0.96 2.62 3.23

Canon Pro 9500Prior art:

CYNSN 27+0 = 27 0.92 2.15 3.03ISglobal-CYNSN 27+9 = 36 0.70 1.86 2.30IS-YNSN 8+12 = 20 0.85 1.94 2.17

new:ISseparately-CYNSN 27+24 = 51 0.54 1.49 2.08ISsingle-CYNSN 27+8 = 35 0.38 0.99 1.40

Table 2.3: Prediction accuracies for both the IS-YNSN model and the IS-CYNSN models whencharacterizing ink spreading using simulated RGB sensors.

Test sets # measurements ∆E94

Model p + i Avg 95% Max

Brother 4000-HLISglobal-CYNSN 27+9 = 36 1.31 3.28 3.77IS-YNSN 8+12 = 20 1.88 4.09 4.93ISseparately-CYNSN 27+24 = 51 1.15 3.29 3.94ISsingle-CYNSN 27+8 = 35 0.99 2.74 3.23

Canon Pro 9500ISglobal-CYNSN 27+9 = 36 0.70 1.92 2.36IS-YNSN 8+12 = 20 0.74 1.71 2.00ISseparately-CYNSN 27+24 = 51 0.57 1.59 2.14ISsingle-CYNSN 27+8 = 35 0.38 1.05 1.47

23


Table 2.4: Prediction accuracies of theCYNSN, IS-YNSN and IS-CYNSN modelsfor the fluorescent and non-fluorescentinkjet ink sets used to establish the fluo-rescent G f printer gamut.

Test sets (inkjet) ∆E94

Model Avg 95% Max

cm f y f

CYNSN 1.41 4.69 6.26IS-YNSN 2.47 5.19 6.60IS-CYNSN 0.86 2.37 3.98

cm f yCYNSN 1.57 4.66 6.73IS-YNSN 1.35 2.82 3.65IS-CYNSN 0.55 1.23 2.90

cmy f


cmykCYNSN 1.33 3.10 5.01IS-YNSN 1.15 2.23 2.77IS-CYNSN 0.44 1.10 1.41

Table 2.5: Prediction accuracies of theCYNSN, IS-YNSN and IS-CYNSN modelsfor the fluorescent and non-fluorescentoffset ink sets used to establish the fluo-rescent G f printer gamut.

Test sets (offset) ∆E94

Model Avg 95% Max

cm f y f


cm f yCYNSN 3.20 9.37 12.04IS-YNSN 1.87 3.81 4.66IS-CYNSN 0.65 1.61 2.83

cmy f


cmykCYNSN 2.33 6.05 7.25IS-YNSN 1.03 2.32 3.32IS-CYNSN 0.49 1.27 1.53

Xrite spectrophotometer with geometry (450 : 00). Tables 2.4 and 2.5 give for respectively

the inkjet and offset prints the mean prediction error in terms of ∆E94 values, the maximal

prediction error and the 95% quantile prediction error. The Yule-Nielsen n-value yielding the

best prediction accuracies for the two printers and for all test sets is 14.

In case of the inkjet and offset prints (Tables 2.4 and 2.5) both the CYNSN and IS-YNSN models

are not accurate for predicting combinations of classical and daylight fluorescent inks with

∆E94 mean prediction errors varying between 1.34 and 3.44, with 95% quantile prediction

errors varying between 2.82 and 9.37, and with a maximal prediction error of 12.84. In contrast,

the IS-CYNSN spectral prediction model is remarkably accurate for predicting combinations

of classical and daylight fluorescent inks. In case of the inkjet prints, the ∆E94 prediction

errors vary between 0.55 and 0.86 with a maximal 95% quantile prediction error of 2.32 for

the cm f y f ink set and in case of the offset prints, the ∆E94 prediction errors vary between

0.65 and 0.97 with a maximal 95% quantile prediction error of 2.71 for the cm f y f ink set. The

IS-CYNSN model also accurately predicts halftones made of classical cmyk inks with a ∆E94

mean prediction error of 0.44 for the inkjet print and of 0.49 for the offset print.

From these prediction accuracies, we conclude that in order to establish the exact fluorescent

sub-gamuts and therefore the fluorescent print gamut G f , we shall use the specially developed

24

2.7. Summary

IS-CYNSN spectral prediction model that is remarkably accurate for predicting combinations

of classical and daylight fluorescent inks. In addition, this model is calibrated with only a few

halftone patch reflectance measurements.

2.7 Summary

We proposed an extension of the cellular Yule-Nielsen spectral Neugebauer model by account-

ing for ink spreading separately within each subdomain. The ink spreading characterization

of the cellular Yule-Nielsen model can be established by jointly fitting the ink spreading in-

terpolation coefficients on a single halftone centered within each subdomain. We obtain

excellent spectral prediction accuracies for predicting halftones combining classical inks only

for inkjet, offset and electrophotography prints. Prediction accuracies are also remarkable

for predicting halftones combining classical and daylight fluorescent inks for both offset and

inkjet prints. Compared with the original cellular Yule-Nielsen model and the ink spreading

enhanced Yule-Nielsen models, prediction accuracies are significantly improved. By jointly

accounting for ink spreading on a single haltone located at the center of the considered cellular

subdomain, we can characterize fluorescent ink halftones with a small number of calibration

measurements. For this reason, in the rest of the thesis, the ink spreading cellular Yule-Nielsen

model (IS-CYNSN) will always refers to this ink spreading characterization. In addition, we

show that ink spreading characterization can be performed with RGB sensors without reduc-

tion of prediction accuracy. This offers the potential of characterizing printers at run time at

moderate cost.

25

3 Framework for printing with com-bined classical and daylight fluores-cent inks

3.1 Introduction

In the previous chapter, we have shown that the IS-CYNSN spectral prediction model accu-

rately predicts the spectral reflectances of hafltones comprising daylight fluorescent inks. This

enables the exact computation of the gamut G f enclosing all colors printable with the 6 ink

print setup. The exact computation of the fluorescent G f gamut is required for establishing

a mapping between the sRGB display gamut GsRGB and the 6 ink printer gamut. In order

to be able to print the displayed colors, we have to establish a mapping between the sRGB

display gamut and the printer gamut (Morovic and Luo 2001). In order to drive the printer,

we also need to establish a relationship between the gamut mapped CIELAB colors and the

corresponding ink surface coverages. For mapping the sRGB gamut into the ink gamut, we

need in a first step to map the lightness range of the sRGB gamut into the lightness range of

the ink gamut by lightness adaptation and to establish the lightness adapted sRGB gamut and

the joint fluorescent gamut of the fluorescent and non-fluorescent ink sets, i.e the conjunction

of the sub-gamuts Gcm f y f , Gcm f y , Gcmy f , Gcmyk . Detailed of the possible lightness adaptation

functions are given in Chapter 4. In the current section, we focus on gamut comparison only.

We are interested to see what is the color domain extent by adding the m f and y f inks to the

classical cmyk inks.

In order to show the colors available by printing combinations of classical and daylight fluo-

rescent inks, we compare for an inkjet printer the Gcmyk and G f gamuts with the sRGB display

gamut GsRGB (Section 3.2). The comparison is done under the A, D65 and F7 illuminants by

calculating the additional volume of the sRGB gamut covered by the 6 ink fluorescent setup,

by calculating the volume of colors offered by the G f gamut and by showing these gamut

boundaries in constant CIELAB lightness planes.

We also compare the different color domain extensions offered by the fluorescent G f gamuts

for both the inkjet and offset printers by showing their gamut boundaries in constant CIELAB

27

Chapter 3. Framework for printing with combined classical and daylight fluorescent inks

lightness planes (Section 3.3).

3.2 Comparison of fluorescent and non-fluorescent ink gamuts

A color device gamut in the CIELAB space is a 3D volume whose surface can be described by

triangles. Every color point inside this surface belongs to the gamut. In order to compute the

lightness adapted sRGB gamut, we generate CIELAB points by varying the sRGB Red, Green and

Blue values by steps of 0.03, convert them to CIE-XYZ tri-stimulus values according to known

phosphor tri-stimulus values, to CIELAB colors and then apply an adaptation of the lightness

L∗ values according to a specific lightness adaptation function, yielding 343 = 39304 CIELAB

color points. For the gamut comparisons, we consider only a non-linear Bézier lightness

adaptation function for mapping the sRGB lightness range to the ink lightness range. This

non-linear function as well as a detailed analysis of different lightness adaptation functions are

given in Chapter 4. In order to obtain the gamut of the 4 considered ink sets, we predict with

the IS-CYNSN spectral prediction model the total reflectance factors of halftones by varying

the nominal ink surface coverages by steps of 0.05 for the 4 ink set cmyk and by steps of 0.03

for the 3 ink sets cm f y f , cm f y , cmy f , convert them to CIE-XYZ tri-stimulus values according

to the spectral power distribution of the considered illuminant and compute CIELAB colors,

yielding 214 + 3 · 343 = 118126 CIELAB color points. The non-convex gamut boundary is

obtained by performing a Delaunay triangulation of the set of CIELAB color points and by

computing with the ball-pivoting technique (Bernardini et al. 1999) the set of surface triangles

defining the concave boundary. The sRGB white is mapped to the paper white by taking as

CIELAB white reference the display white for converting sRGB values to CIELAB colors and

the paper white for converting tri-stimulus values of print samples from CIE-XYZ to CIELAB

colors.

In order to show the new colors available by printing combinations of classical and daylight

fluorescent inks, we establish for the inkjet Canon Pro 9500 printer the non-fluorescent Gcmyk

gamut comprising only the colors generated with the classical cmyk inks, the joint fluorescent

G f gamut comprising colors generated with the classical inks and the daylight fluorescent

magenta and yellow inks, and the non-linearly lightness adapted sRGB gamut G ′sRGB .

Under the D65 illuminant, Figure 3.1a shows in 3D the Gcmyk and the G f gamuts. At high

lightness values (L∗ > 60), we observe a gamut extension in the yellow, magenta, red and green

parts of the gamut due to the fluorescence of the y f and m f inks. Figure 3.1b illustrates a

comparison between the joint fluorescent G f and the non-linearly lightness adapted G ′sRGB

gamuts. A significant part of the G f gamut is outside the G ′sRGB gamut. In Chapter 4, we

propose a suitable mapping from the G ′sRGB gamut to the G f gamut that enables expanding

the input G ′sRGB gamut colors into the G f printable gamut.

Let us compare the gamuts within constant lightness planes. Figure 3.2 illustrates a com-

parison between the non-linearly lightness adapted G ′sRGB gamut, the Gcmyk gamut and the

fluorescent gamut G f under the D65 illuminant. Gamut boundaries are shown within constant

28

3.2. Comparison of fluorescent and non-fluorescent ink gamuts

(a) (b)a*

100

80

60

40

-40 -20 0 20 40 60 -40 0 40

b*

L*

Gf

Gcmyk

100

80

60

40

L*

-40 -20 0 20 40 60a*

-40 0 40

b*

Gf

G'sRGB

Figure 3.1: (a) Comparison between the gamut Gcmyk (colored solid) and the joint fluorescentink gamut G f (mesh grid) and (b) comparison between the joint fluorescent gamut G f (coloredsolid) and the display non-linearly lightness adapted gamut G ′

sRGB (mesh grid).

lightness planes from L∗ = 50 to L∗ = 95.

At a lightness between L∗ = 55 and L∗ = 65, there are not many differences between the

classical ink gamut Gcmyk and the joint fluorescent gamut G f . For lightnesses lower than

L∗ = 50, there is strictly no difference between the Gcmyk and the G f gamuts. For lightnesses

between L∗ = 70 and L∗ = 80, the G ′sRGB gamut is significantly better covered by the joint

fluorescent gamut G f in the green, orange, magenta, yellow and blue parts of the gamut than

by the classical ink gamut Gcmyk . This is due to the high saturation offered by the daylight

fluorescent inks, as become apparent in the spectral reflectances of the m f and y f inks (Figures

1.2a and 1.2b, pointed lines). For lightnesses varying between L∗ = 85 and L∗ = 95, the classical

inks fill only a small part of the sRGB gamut while the joint fluorescent gamut covers most of

the sRGB gamut. At L∗ > 95, only the G f gamut fills the yellow and green parts of the G ′sRGB

gamut.

For establishing the G f gamut under the F7 illuminant, fluorescent and non-fluorescent ink

set halftones were measured under the F7 illuminant by making use of the Just Normlicht mini

5000 light table as light source, by building a custom measurement setup with geometry (00 :

450) comprising an optical fiber capturing the reflected light, connected with a Maya Pro 2000

spectrophotometer. In case of establishing the G f gamut under the A illuminant, the halftones

were measured with the A illuminant of a SpectroEye Xrite spectrophotometer with geometry

(450 : 00). CIELAB color points defining the gamut are predicted with the IS-CYNSN spectral

prediction model. Accuracies of the IS-CYNSN model for the fluorescent and non-fluorescent

ink sets under these two illuminants are listed in Table 3.1. For measurements performed

under the A illuminant, we obtain for the cmyk and cm f y f test sets remarkable prediction

accuracies, with respective mean∆E94 prediction error of 0.34 and 0.49 and respective quantile

95% prediction error of 0.80 and 1.10. In the case of measurements performed with the

Just Normlich light table (illuminant F7), we obtain slightly less accurate predictions with

29


-50

0

50

-50 0 50a*

-50 0 50a*

L* = 75

-50

0

50

L* = 55

-50 0 50a*

G'sRGB

Gcmyk

Gf

-50

0

50

b*b*

L* = 65

L* = 90

-50 0 50a*

-50

0

50

-50

0

50

b*b*

-50 0 50a*

L* = 95

-50

0

50

b*b*

-50 0 50a*

L* = 85

Figure 3.2: Color gamuts of the non-linearly lightness adapted sRGB space (solid lines), the4 ink print gamut using the classical cyan, magenta and yellow inks (dashed lines) and thejoint fluorescent gamut using the classical cyan, magenta, yellow, black and the two additionaldaylight fluorescent magenta and yellow inks (dotted lines) under the D65 illuminant.

a mean ∆E94 prediction error of 1.10 when predicting the 625 classical cmyk test samples

and of 0.71 when predicting the 125 cm f y f test samples. Since the test sets containing

one daylight fluorescent ink (cm f y and cmy f ) show the same prediction accuracies as the

set containing two daylight fluorescent ink (cm f y f ) they are not listed in Table 3.1. These

prediction accuracies enable establishing the exact G f gamuts under both the A and D65

illuminants.

Figure 3.3 compares the joint fluorescent G f gamuts under the D65, A and F illuminants and

the non-linearly lightness adapted G ′sRGB gamut. Due to its low energy within the UV and

blue excitation wavelength ranges, the A illuminant induces less fluorescence (Figure 1.2,

dashed lines) and therefore provides the smaller gamut (GA dashed lines). The D65 illuminant

provides the largest fluorescent ink gamut included within the G ′sRGB gamut, especially in

the green, magenta and yellow parts. The F7 illuminant provides also a strong fluorescent

emission of the daylight fluorescent colorants, but achieve less coverage of the G ′sRGB gamut,

compared with the D65 illuminant.

30

3.2. Comparison of fluorescent and non-fluorescent ink gamuts

Table 3.1: Prediction accuracies of the IS-CYNSN model for 625 cmyk and 125 cm f y f testsamples printed with a Canon Pro 9500 inkjet printer and measured under the A and F7illuminants.

Illuminant ∆E94

Test set Avg 95% Max

Xrite Acmyk 0.34 0.80 1.23cm f y f 0.49 1.10 2.58

Just Norm. F7cmyk 1.10 2.89 5.38cm f y f 0.71 2.11 3.57

G'sRGB

G F7

G D65G A-50

0

50

b*

-50 0 50a*

L* = 55 L* = 65

-50 0 50a*

-50

0

50

b* b*

L* = 85

-50

0

50

b*

-50 0 50a*

b*

L* = 90

-50

0

50

-50 0 50a*

b*

L* = 75

-50 0 50a*

-50

0

50

L* = 95

-50

0

50

-50 0 50a*

Figure 3.3: Comparison of the non-linearly lightness adapted sRGB gamut G ′sRGB (thick solid

lines) and the fluorescent 6 ink gamuts under the D65 (GD65, solid lines), F7 (GF7 , dotted lines)and A (GA, dashed lines) illuminants.

3.2.1 Comparison of gamut volumes

In this section, we compare the gamut of the classical cmyk print setup with the joint flu-

orescent ink gamut by showing the percentage of the sRGB gamut that can be reproduced.

This can be done by computing the volume of the intersection of the G ′sRGB gamut with the

31


print gamuts. Bala and Dalal (1997) proposed to compute the volume of a printer gamut by

summing the volume of tetrahedra formed by the vertices of each surface triangle defining its

boundary and a point inside the gamut. With the volume of the non-linearly lightness adapted

sRGB gamut V (G ′sRGB ), the volume of the joint fluorescent gamut V (G f ) and the volume of

the conjunction of the G f and G ′sRGB gamuts, we can compute the volume of the intersection

of the G f gamut with the G ′sRGB gamut with the following equation

V (G f ∩G ′sRGB ) =V (G f )+V (G ′

sRGB )−V (G f ∪G ′sRGB ) (3.1)

In an analog manner, we can compute the intersection of the Gcmyk and G f gamut with the

G ′sRGB gamut.

Table 3.2 shows the volumes of the classical Gcmyk and joint fluorescent G f ink gamuts, and

the volumes formed by overlap between their gamuts and the G ′sRGB gamut in thousands

L∗a∗b∗ volume units under the A, D65 and F7 illuminants. It also shows the gain in overlapped

volume offered by the joint fluorescent 6 ink gamut, in respect to the traditional cmyk gamut.

It further shows the gain gsr relative to the volume of the non-linearly lightness adapted sRGB

gamut G ′sRGB , as defined by Equation 3.2:

gsr =V (G f ∩G ′

sRGB )−V (Gcmyk ∩G ′sRGB )

V (G ′sRGB )

(3.2)

Table 3.2: Comparison of gamut volumes for the G ′sRGB gamut.

Illuminant D65 A F7

Volume Gcmyk 288 282 283Volume G f 451 356 417Volume Gcmyk ∩G ′

sRGB 286 241 254Volume G f ∩G ′

sRGB 373 276 328gain 31% 14% 29%gsr gain relative to sRGB′ volume 13% 5% 11%(V of G ′

sRGB = 693)

Since the A illuminant has less energy in the fluorescent excitation range, the A illuminant

yields the smallest gamut volume. By comparing the classical 4 ink gamut with the joint

fluorescent 6 ink gamut, we observe a gamut volume increase of 31% and 29% for respectively

the D65 and F7 illuminants. When considering the part of the G ′sRGB gamut covered by

respectively the joint fluorescent and classical ink gamuts, under the D65 illuminant, 54% of

the G ′sRGB gamut are covered instead of 41% and under the F7 illuminant 47% of the G ′

sRGB

gamut are covered instead of 36%. Since the CIELAB color space is not perceptually constant

32

3.3. Comparison of inkjet and offset fluorescent ink gamuts

for large color differences, these computed volumes are approximate. A better approximation

of these volumes may be obtained with a more perceptually constant color space, such for

instance the LAB2000HL color space (Lissner and Urban 2012).

L* = 75L* = 55 L* = 65

L* = 90 L* = 95L* = 85

-50

0

50

b*

-50 0 50a*

-50 0 50a*

-50

0

50

b*

-50

0

50

b*

-50 0 50a*

-50

0

50

b*

-50 0 50a*

-50

0

50

b*

-50 0 50a*

-50 0 50a*

-50

0

50

b*

Figure 3.4: Comparison of the inkjet (solid lines) and offset (dashed lines) fluorescent G f

gamuts under the D65 illuminant.

3.3 Comparison of inkjet and offset fluorescent ink gamuts

In this section, we compare the fluorescent printer gamuts obtained by printing combinations

of classical and daylight fluorescent inks either with inkjet inks or with offset inks. Figure 3.4

illustrates this comparison as CIELAB constant lightness planes from L∗ = 55 to L∗ = 95. For

lightnesses between L∗ = 55 and L∗ = 75, the offset G f gamut shows larger boundaries in the

red and orange parts of the gamut. These larger boundaries are due to the more saturated red

colorant obtained by superposing the offset m f and y f inks than by superposing the inkjet m f

and y f inks (Figure 1.3c). Since the offset m f colorant has a larger fluorescent emission peak

near 600 nm with a reflectance factor of 1.31 than the inkjet m f colorant with a reflectance

factor of 1.20 (Figure 1.3a), we also observe for the offset G f gamut larger boundaries in the

magenta part of the gamut. The inkjet G f gamut shows for these lightnesses larger boundaries

in the green regions of the gamut. Since the y f inkjet colorant is more saturated than the y f

offset colorant (Figure 1.3b), we therefore obtain a more saturated fluorescent green colorant

33


by superposing the inkjet cyan and y f inks than by superposing the offset cyan and y f inks.

For lightnesses more than L∗ = 75, while the inkjet G f gamut shows larger boundaries in

the green region of the gamut, the offset G f gamut shows larger boundaries in the orange

and yellow parts of the gamut. Since within the green visible wavelength range near 520 nm,

the inkjet y f colorant has a significant higher fluorescent emission peak than the offset y f

colorant with respective reflectance factors of 1.41 and 0.92 (Figure 1.3b), the inkjet domain

extension of colors for high lightnesses is mainly visible in the green region. For the offset y f

colorant, the two fluorescent peaks are located in both the green and red wavelength range,

yielding strong fluorescent yellow colors.

3.4 Summary

Thanks to the fluorescent ink halftones accurate spectral predictions of the IS-CYNSN spectral

prediction model, we establish for the inkjet printer the fluorescent gamut G f combining

the classical cyan, magenta, yellow and black inks together with the two additional daylight

fluorescent yellow and magenta inks. We compute the classical and fluorescent gamut volumes

and compare their gamut volumes in respect to the sRGB gamut, under the D65, fluorescent

F7 and A illuminants. The large inkjet fluorescent gamut is present for light source having

much energy in the UV and blue wavelength range, such as the D65 and F7 illuminants. By

comparing the boundaries of the fluorescent gamut with the boundaries of the lightness

adapted sRGB display gamut, under the D65 and F7 illuminants the fluorescent gamut covers

at least 11% more the sRGB display gamut, compared with the classical cmyk gamut. From the

computation of the gamut volumes and the comparison of gamut boundaries, we conclude

that the two additional inkjet daylight fluorescent yellow and magenta inks considerably

expand the gamut of classical printers for images that are viewed under normal daylight

and fluorescent tubes. The fluorescent gamuts extend the domain of colors mainly at high

lightness from L∗ = 55 to L∗ = 100.

We also compare the fluorescent gamuts G f obtained with offset and inkjet printers. The

offset printer shows a color domain extension larger in the magenta, orange and yellow parts

of the gamut than the one obtained with the inkjet printer, while the inkjet printer shows a

color domain extension larger in the green part of the gamuts than the one obtained with

the offset printer. In the next part part of this thesis, we use these new fluorescent colors to

both better reproduce input sRGB colors and to highlight image regions of special interest by

making use of the new available bright and high chroma fluorescent green, yellow, orange, red

and magenta colors.

34

4 Gamut mapping expansion and re-duction for color reproduction withdaylight fluorescent inks

4.1 Introduction

In this chapter, we show how to map input sRGB images to printable destination fluorescent

images by gamut expansion and reduction mapping algorithms (Section 4.3). Gamut mapping

contraction and expansion techniques are not new (Morovic and Luo 2001), (Kang et al. 2003).

However, we adapt these techniques in order to accommodate the destination fluorescent ink

gamut G f .

We also have the goal of enhancing specific image parts by printing them with high chroma

and bright colors. For this purpose, we developed a software that enables controlling the

gamut expansion. We first select the image parts to be enhanced. We then apply to their colors

a user-controlled gamut expansion that increases both their chroma and lightness towards

colors located at the boundary of the destination fluorescent ink gamut G f . We also create

smooth chroma transitions between the expanded and non-expanded image parts (Section

4.4). We then preview the resulting printable gamut expanded image (Section 4.5). Finally, we

show how to perform the color separation into its 6 ink color separation layers (Section 4.6).

The resulting prototype software enables artists to create and print their own designs.

4.2 Mapping the lightness range of the sRGB gamut into the ink des-

tination gamut

Mapping the sRGB gamut into the ink gamut requires in a first step to map the lightness

range of the sRGB gamut into the lightness range of the ink gamut (Morovic and Luo 2001). In

order to map the GsRGB lightness range into the destination gamut lightness range, we first

determine the minimal lightness L∗inksMin of the inks, i.e. the lightness of the solid pure black

ink. We may then either apply a linear mapping that better preserve lightness differences of

the input sRGB image space but raises all sRGB lightnesses, apply a partly non-linear mapping

35

Chapter 4. Gamut mapping expansion and reduction for color reproduction withdaylight fluorescent inks

40

0

20

60

80

100

L*inksMin

L* inks

6020 40 80 100L*

sRGB

Linear

P1

P2

Figure 4.1: Linear (dashed line) and non-linear mapping of sRGB lightnesses. For the non-linear mapping, the input sRGB lightness values are either non-linearly mapped betweenL∗ = 0 and L∗ = 50 and preserved beyond L∗ = 50 (solid line) or preserved between L∗ = 50and L∗ = 60 and shifted to higher values beyond L∗ = 60 (pointed line).

that preserves high lightness values but maps low lightness values into a smaller lightness

range or apply an s-shape like non-linear mapping. These lightness mappings can be defined

with a cubic Bézier function

B(t ) = (1− t )3 ·P0 +3 · (1− t )2 · t ·P1 +3 · (1− t ) · t 2 ·P2 + t 3 ·P3 (4.1)

where P0, P4 are the points (0,L∗inksMin), (100,100), and where parameter t is varied between

zero and one.

Control points P1 and P2 are user-defined according to the type of mapping. Figure 4.1

illustrates a typical case of a sRGB lightness adaptation for an inkjet Epson P50 printer which

has a minimal L∗inksMin at 23. In case of a linear lightness mapping, the control points P1 and

P2 are respectively set to (0,L∗inksMin) and (100,100) (Figure 4.1, solid line). In case of a partly

non-linear mapping, P1 and P2 are both set to (L∗inksMin,L∗

inksMin). This preserves lightnesses

for L∗ > 50 (Figure 4.1, dashed line). In case of a s-shaped non-linear mapping, lightnesses

between L∗ = 50 and L∗ = 60 are preserved, but raised at lightnesses L∗ > 60. Control points

P1 and P2 can for instance be set to (2 ·L∗inksMin,L∗

inksMin) and (80,100) (Figure 4.1, pointed

line).

By comparing in 3D the non-linearly lightness adapted sRGB gamut where control points P1

and P2 are both set to (L∗inksMin,L∗

inksMin) and the linearly lightness adapted sRGB gamut where

control points P1 and P2 are respectively set to (0,L∗inksMin) and (100,100) with the G f gamut

(Figure 4.1), we observe that the linearly adapted sRGB gamut G ′′sRGB achieves more overlap

with the fluorescent ink gamut G f for lightnesses L∗ > 70. However, for lower lightnesses it

achieves less overlap than the non-linearly lightness adapted gamut G ′sRGB , especially in the

36

4.2. Mapping the lightness range of the sRGB gamut into the ink destination gamut

(a)

100

80

60

40

-40 -20 0 20 40 60a*

-40 0 40

b*

100

80

60

40

-40 -20 0 20 40 60a*

-40 0 40

b*

L* L*

G'sRGB

GfGf

G''sRGB

(b)

Figure 4.2: (a) Comparison between the fluorescent ink gamut G f (colored solid) and thedisplay linearly lightness adapted gamut G ′′

sRGB (mesh grid) and (b) comparison between thefluorescent ink gamut G f (colored solid) and the non-linearly lightness adapted display gamutG ′

sRGB (mesh grid).

Table 4.1: Comparison of gamut volumes for the G ′′sRGB gamut.

Illuminant D65 A F7

Volume Gcmyk ∩G ′′sRGB 271 243 247

Volume G f ∩G ′′sRGB 368 285 323

gain 36% 17% 31%gsr gain relative to sRGB′′ volume 16% 7% 13%(V of G ′′

sRGB = 598)

blue part of the gamuts. This linearly lightness adapted G ′′sRGB gamut has colors shifted to

slightly higher lightness values. The G f gamut therefore covers a larger fraction of the G ′′sRGB

gamut. Under the D65 and F7 illuminants, respectively 62% and 54% of the G ′′sRGB gamut are

covered (See Table 4.1) instead of 54% and 47% of the G ′sRGB gamut (See Table 3.2, Chapter 3,

Section 3.2.1).

This Bézier lightness adaptation function can be user-defined depending on the darkest

color achievable by a given set of inks and printer and also on the input image content. For

bright images, mapping a large range of dark colors into a small range of dark colors enables

preserving the lightnesses of the original image without loss of details within the printed

image. On the contrary, for dark images, it is preferable to linearly map the lightnesses with

the goal of preserving the input image details in dark tones.

37


L*

C *

L*=lh

L*=ll Gsource

Gdest

CsourceMax

CdestMax

Csource

Cdestα⋅CsourceMax

I

II

(a)

L*

C *

L*=lh

L*=ll Gsource

Gdest

CsourceMax

CdestMaxα⋅CdestMaxII

CdestCsourceI

(b)

Figure 4.3: Multiple foci gamut mapping approach for (a) gamut expansion of point Csource toCdest and (b) gamut reduction of point Csource to Cdest .

4.3 Mapping the lightness adapted sRGB gamut onto the printable

fluorescent gamut

In order to map the sRGB gamut onto the printable destination gamut, we first establish the

lightness adapted sRGB gamut G ′sRGB by varying the sRGB Red, Green and Blue values in small

steps, convert them to CIE-XYZ values and then to CIELAB. We then apply an adaptation of the

lightness L∗ values according to the desired lightness range mapping function with Equation

4.1 and establish the concave gamut by Delaunay triangulation and ball-pivoting (Bernardini

et al. 1999). The fluorescent gamut is established by predicting with the IS-CYNSN spectral

prediction model the total reflectance factors of halftones by varying nominal ink surface

coverages by small steps for the 4 ink sets cmyk, cm f y f , cm f y , cmy f , by converting them to

CIE-XYZ tri-stimulus values according to the D65 illuminant, by computing corresponding

CIELAB colors and by deriving with ball-pivoting the set of surface triangles defining the

concave boundary. The sRGB white is mapped to the paper white by taking as CIELAB white

reference the display white for converting sRGB values to CIELAB colors and the paper white

for converting tri-stimulus values of print samples from CIE-XYZ to CIELAB colors.

The lightness adapted gamut G ′sRGB is mapped to the printable destination fluorescent gamut

according to a multiple foci approach (Morovic and Luo 2001), as shown in Figure 4.3. We

define an upper lightness bound lh and a lower lightness bound ll . For a color point Csource

of the source gamut Gsource, we apply a mapping from point Csource to Cdest on a line passing

through Csource and through the focal point on the black and white axis for Csource color points

that have a lightness either Lsource > lh or Lsource < ll or on a constant lightness line passing

through Csource when ll ≤ Lsource ≤ lh (Figure 4.3, pointed lines). The mapping line intersects

the source and destination gamut boundaries at respective intersection points CsourceMax and

CdestMax.

38

4.3. Mapping the lightness adapted sRGB gamut onto the printable fluorescent gamut

In case of gamut expansion, i.e. CdestMax > CsourceMax (Figure 4.3a), we apply a chroma ex-

pansion to chroma values Cdest by mapping the interval (I): [α ·CsourceMax,CsourceMax] into the

interval (II): [α ·CsourceMax,CdestMax] according to the following equation

Cdest =α ·CsourceMax +(

Csource −α ·CsourceMax

CsourceMax −α ·CsourceMax

)γ· (CdestMax −α ·CsourceMax) (4.2)

where factor γ expresses a possible non-linearity of the chroma mapping. With γ = 1, the

mapping is linear and with 0 < γ< 1, chroma is non-linearly expanded. Factor α between 0

and 1, defines the internal part of the source gamut where chroma values do not change, i.e.

within the interval 0 ≤α ·CsourceMax we have Cdest =Csource. This prevents the chroma increase

of low chroma colors.

Figure 4.4 illustrates the need of a multiple foci approach for expanding color points. In

case of chroma expansion towards the color of a daylight fluorescent colorant, we should

avoid mapping a small range of colors into a large range of colors. For the considered inkjet

printable fluorescent gamut, we set the upper bound to lh = 76. In the hue plane of the inkjet

m f colorant, compared with constant lightness mapping (Figure 4.4a, upper dashed line) the

non-constant lightness mapping line passing through the inkjet m f ink color (Figure 4.4a,

upper pointed line) maps a significantly larger range of input gamut colors. The non-constant

lightness mapping is also required in order to map input colors into the color of the inkjet

y f colorant whose maximal lightness is L∗ = 108, i.e. higher than the maximal source gamut

lightness at L∗ = 100. For mapping inkjet green colors, e.g. at a hue angle of 1470, applying a

constant lightness mapping in the lightness range ll ≤ L∗ ≤ lh = 76 is appropriate since the

source range of color is large enough in comparison with the destination gamut (Figure 4.4b,

dashed line). In addition, the upper lightness focal point also controls the lightness shifts when

expanding source colors. For instance, by decreasing the lightness of the focal point lh , colors

with lightness higher than lh are expanded towards higher lightnesses. Other hue planes, from

a hue angle of 0 degree to 300 degrees by steps of 60 degrees are shown in Appendix A.

Figure 4.5 illustrates the need of a multiple foci approach for expanding color points for the

considered offset printer. We set the upper lightness bound lh to 85. When expanding colors

towards the color of the offset fluo y f colorant, the non-constant mapping line passing through

the offset y f colorant maps a significant large range of colors (Figure 4.5a, upper pointed line).

In case of mapping colors towards the red fluo colorant, constant mapping is appropriate

since the destination color range is large enough in comparison with the source gamut (Figure

4.5b, dashed line). We also show a comparison between the non-linearly lightness adapted

G ′sRGB gamut and the offset fluorescent gamut G f as constant hue planes from 0 degree to 300

degrees in Appendix A.

In case of gamut reduction, i.e. CsourceMax ≥CdestMax (Figure 4.4b), we apply a chroma reduc-

tion to chroma values Cdest by mapping the interval (I): [α ·CdestMax,CdestMax] according to the

39


C *

20 40 60 800

(a)

20

40

60

80

100

L*

hue angle of 352°Gf

G'sRGB

mf

lh

20

40

60

80

100

L*

20 40 60 80 1000

C *

(b)

gf

Gf

G'sRGB

hue angle of 146°

lh

ll ll

c1 c2

Figure 4.4: Constant hue planes for an inkjet Epson P50 printer (a) at a hue angle of 3520

showing both a constant (upper dashed line) and a non-constant (upper pointed line) ex-pansion lightness mapping line in direction of the magenta fluo solid ink (m f ), and botha constant (lower dashed line) and non-constant (lower pointed line) reduction lightnessmapping line for a low lightness color (c1) and (b) at a hue angle of 1460 showing a constantexpansion lightness mapping line (dashed line) in direction of the green fluo colorant (g f ) andnon-constant reduction lightness mapping line (pointed line) for a low lightness color (c2).

following equation

Cdest =α ·CdestMax +(

Csource −α ·CdestMax

CsourceMax −α ·CdestMax

)· (CdestMax −α ·CdestMax) (4.3)

The lower lightness bound ll used for reducing chroma prevents a too strong chroma reduction

in dark tones and therefore better preserves the original image colors. In case of the inkjet

printer, we set ll to 40. When reducing chroma of a low lightness color c1 located in a hue plane

at 3520, the non-constant lightness mapping line passing through c1 and a single point in the

black and white axis at a lightness L∗ = 40 (Figure 4.4a, lower pointed line) reduces less chroma

than a constant lightness mapping line (Figure 4.4a, lower dashed line). The non-constant

lightness mapping line for reducing chroma in dark tones, e.g. a dark c2 color (Figure 4.4b) is

not required in a hue plane at 1460 since the destination gamut G f boundary is as large as the

source gamut G ′sRGB . However, in order to ensure color continuity while reducing chroma, we

use the same low lightness ll bound for all hue planes. For the considered offset printer, we

set the lower lightness bound ll to 30. In this situation, the chroma of a low lightness color c3

located in a hue plane at an angle of 980 is not reduced too much (Figure 4.5a, lower pointed

line).

40

4.4. User driven gamut expansion

20

40

60

80

100

L*

hue angle of 98°

20 40 60 800

C *

20

40

60

80

100

C *20 40 60 800 100

L*

hue angle of 41°

(a) (b)

lh lh

c3

ll

yf

rf

13100

13

Gf

G'sRGBG'sRGB

Gf

Figure 4.5: Constant hue planes for an offset Heidelberg printer (a) at a hue angle of 980

showing a non-constant (upper pointed line) expansion lightness mapping line in directionof the yellow fluo solid ink (y f ), and a non-constant (lower pointed line) reduction lightnessmapping line for a low lightness color (c3) and (b) at a hue angle of 410 showing a constantlightness mapping line (dashed line) in direction of the red fluo colorant (r f ).

4.4 User driven gamut expansion

In this section, we define user parameters that enable controlling the gamut expansion. We

would like to control the chroma enhancement within image parts selected by users. Within

image parts, we apply a gamut expansion of the input image sRGB colors by mapping the input

chroma interval either linearly or non-linearly onto the printable output chroma interval, by

considering the maximal chroma or a part of the maximal chroma achievable by the printable

fluorescent gamut. In addition, since within the select image part we may strongly increase

both chroma and brightness of the colors, it is possible to see strong color differences at the

boundary between the expanded and non-expanded image parts. In order to reduce these

strong color differences, we create smooth chroma transitions from the center of the selected

image part reproduced with high chroma colors to the border of the selection reproduced

with lower chroma colors. Input sRGB image colors are mapped according to the equations

presented in Section 4.3 with additional user-defined parameters that are described below.

Outside the selected image parts, no sRGB gamut expansion is performed. An input Csource

color is mapped according to Equation 4.3 when CdestMax ≤ CsourceMax or when CdestMax >CsourceMax is kept as it is, i.e. Cdest = Csource. Within the selected image parts, the lightness

adapted sRGB chroma can be enhanced. We distinguish two cases. The first case is when

the destination fluorescent gamut is greater than the source gamut along the mapping line

(CdestMax > CsourceMax). In this case, the chroma can be expanded according to Equation

4.2 and possibly with the non-linear chroma reinforcement factor γ varying within 0 < γ<1. This chroma expansion can be limited according to a user-defined chroma expansion

limitation factor δ. This chroma expansion limitation factor limits the effective maximal

41


chroma expansion CdestMaximalExp to values between CdestMax and CsourceMax. The effective

maximal chroma expansion then becomes

CdestMaximalExp =CsourceMax ·δ+CdestMax · (1−δ) (4.4)

and replaces CdestMax in Equation 4.2, for 0 ≤ δ ≤ 1. Parts of images reproduced with the

maximal expansion limitation factor δ= 1 do not contain chroma expanded colors.

The second case is when the source gamut is greater than the destination gamut along the

mapping line (CsourceMax > CdestMax). In this case, the chroma of a Csource color can be rein-

forced by non-linearly increasing the source Csource chroma towards the chroma CsourceMax

with the non-linear chroma reinforcement factor γ as follows

CsourceExp =α ·CdestMax +(

Csource −α ·CdestMax

CsourceMax −α ·CdestMax

)γ· (CsourceMax −α ·CdestMax) (4.5)

and then performing the gamut reduction CsourceExp to Cdest according to Equation 4.3 by re-

placing Csource with CsourceExp. This yields a reinforced chroma color Cdest within the printable

fluorescent gamut. By using the same chroma reinforcement factor γ in both the chroma

expandable and the non-expandable parts of the input gamut, we ensure the continuity of

the mapped colors. The two user parameters δ and γ respectively limit the maximal gamut

expansion and provide a non-linear increase of the chroma.

In order to suppress strong chroma differences at the boundaries between selected and non-

selected image parts, we create smooth chroma transitions at the proximity of the boundaries

of the selected image parts. For this purpose, we establish a spatial interpolation map with

values varying between 1 and 0. The final colors are obtained by interpolation between the

gamut mapped colors CdestExp with user-defined γ and δ parameters and the non-expanded

destination colors CdestNonExp located outside the selected image parts.

C ′dest =CdestExp ·∆(x, y)+CdestNonExp ·

(1−∆(x, y)

)(4.6)

where the ∆ values are given by the spatially laid out interpolation map. With ∆ = 1, C ′dest

represents the used-defined gamut expanded colors and with ∆= 0, C ′dest represents the non-

expanded colors located outside the selected image parts. The spatial interpolation map is

created with the distance transform algorithm (Rosenfeld and Pfaltz 1968). Pixels outside the

user selected image part are set to black and inside the selection to white. We then apply the

distance transform to obtain for each white pixel its distance to the nearest black pixel. This

42

4.5. Display preview

Figure 4.6: Spatial interpolation map for an arbitrary selection (red line) generated with adistance limitation factor κ= 2.5, where white represents 1 and black represents ;.

distance map is normalized by dividing its values by its maximal value. In order to limit the

distance from the boundary where the interpolation is performed, we multiply the map with a

distance limitation factor κ (1 < κ). Values of the map greater than one are set to 1. Figure 4.6

shows the generated spatially laid out interpolation map for an arbitrary selection (red line)

when using a distance limitation factor κ= 2.5.

By spatially interpolating colors between non-expanded colors CdestNonExp and expanded

colors CdestExp, we create smooth chroma transitions along the boundaries of the selected

image parts.

4.5 Display preview

We developed a tool for designers enabling selecting image parts, applying to these selec-

tions the user-defined gamut expansion parameters described in the previous section and

previewing the printable destination gamut expanded image. In order to display accurate

colors we have to characterize the display device. Display characterization is performed by

computing its gamma exponent γlum correction and by computing the 9 coefficient matrix

used to convert CIE-XYZ tri-stimulus values to linear RGB phosphor values (Brainard et al.

2002). In order to display the chroma expanded image parts, we need to show on a sRGB

display colors located beyond the original display gamut. With the goal of both preserving

the overall appearance of the destination image and observing the differences between the

expanded and non expanded image colors, we simulate on a standard sRGB display a display

having lower tri-stimulus phosphor values. This simulated lower luminance display renders

the non-expanded colors as well as the expanded colors.

We characterized both the Dell U2212 HM and the Eizo ColorGraphic CG245W displays. Their

gamma exponent γlum correction has been obtained by displaying 11 gray patches, i.e. by

using for the three Red, Green and Blue channels 11 evenly distributed values between 0

and 1. Here we assume that sRGB Red, Green and Blue values vary between 0 and 1. The

emitted irrandiances of these 11 gray patches have been measured with a Maya Pro 2000

43


50 1000

50

100

Input luminance Y

Out

put l

umin

ance

Y

(a)

γlum= 1.89

50 1000

50

100

Input luminance Y

Out

put l

umin

ance

Y

(b)

γlum= 2.24

Figure 4.7: Luminance gamma curves γlum for (a) the Dell U2212 HM display and (b) for theEizo ColorGraphic CG245W display. Solid lines show the gamma curves γlum that approximatethe measured CIE-XYZ luminance Y channels (black circles).

spectrophotometer and then converted to CIE-XYZ tri-stimulus values. We then fitted the

gamma curve by minimizing the differences between the measured CIE-XYZ luminance Y

channels with the predicted CIE-XYZ luminance Y channels. Since the gamma curves for

the three Red, Green and Blue phosphors are identical for these two displays, we fit only one

gamma curve. We obtained for the Dell U2212 HM display a luminance gamma curve with

γlum = 1.91 and for the Eizo ColorGraphic CG245W a luminance gamma curve with γlum = 2.24.

Note that the Eizo ColorGraphic CG245W display is a quality monitor calibrated from the

factory. It therefore shows an ideal gamma curve γlum close to 2.2. Figure 4.7 shows the γlum

curves (black lines) that approximates the measured CIE-XYZ Y luminance channel (black

circles) for (a) the Dell U2212 HM display and for (b) the Eizo ColorGraphic CG245W display.

For computing the 9 coefficients a1 to a9 of the matrix converting CIE-XYZ tri-stimulus values

to RGB linear values [Rl Gl Bl ], we measured the emitted irrandiances of the displayed maximal

Red, Green and Blue sRGB phosophor values, i.e the sRGB [RsGsBs] respective component

values [1 0 0], [0 1 0], [0 0 1]. The nine a1 to a9 unknowns are found by solving the following

equation system for these three maximal Red, Green and Blue sRGB component values, i.e.

each sRGB value yields 3 equations.

[Rl Gl Bl ] = [RsGsBs]γlum

[Rl Gl Bl ] = [X Y Z ] · ~M

where ~M =

a1 · · · a3...

. . ....

a7 · · · a9

(4.7)

44

4.5. Display preview

In order to test the characterization of these two displays, we displayed 125 uniformly dis-

tributed patches by varying the sRGB Red Rs , Green Gs and Blue Bs phosphor values by step

of 25% (53 = 125). The CIE-XYZ tri-stimulus values of these displayed patches are either

measured or predicted according to the following equation derived from Equation 4.7

[X Y Z ] = [RsGsBs]γlum · (~M)−1 (4.8)

We finally converted both the predicted and measured CIE-XYZ tri-stimulus values to CIELAB

colors, with the display white acting as reference white. The average ∆E94 prediction error, the

quantile 95% prediction error and the maximal prediction error for both the two considered

displays are listed in Table 4.2. We obtain excellent prediction accuracies with a mean ∆E94

prediction error of 1.16 for the Dell display and of 0.66 for the Eizo display. The Eizo display is

remarkable stable yielding a quantile 95% prediction error of ∆E94 = 1.59.

Table 4.2: Prediction accuracies for both the Dell U2212 HM and the Eizo ColorGraphicCG245W display characterizations.

Display ∆E94

Avg 95% Max

Dell U2212 HM 1.16 3.64 4.87Eizo ColorGraphic CG245W 0.66 1.59 3.89

Once the display is characterized, printable gamut mapped CIELAB values that are to be

previewed are transformed into CIE-XYZ tri-stimulus values X Y Zprintable, using the measured

CIE-XYZ tri-stimulus display white as reference. We then apply to the X Y Zprintable values a

multiplicative tri-stimulus reduction factor ε (0 ≤ ε ≤ 1) and obtain the simulated X Y Zsim

values. The simulated X Y Zsim values are then transformed to display sRGB values according to

Equation 4.8 with the display characterized by its gamma value γlum and by the nine coefficient

matrix.

The system (or the user) can modify the tri-stimulus reduction factor ε until no displayed

values saturates the sRGB display Red, Green and Blue channels. By assuming that the eye

adapts on the simulated ε · XnYn Zn white reference, it becomes possible to visualize the

selected gamut expanded image parts. Since the maximal lightness, respectively maximal

chroma obtainable with the 6 ink print setup (offset or inkjet printers) is L∗ = 108, respectively

C∗ = 123, values of the tri-stimulus reduction factor ε are always above 0.7.

45


4.6 Halftoning and printing

For printing an input sRGB image mapped according to user-defined gamut mapping pa-

rameters, we generate the 6 ink separation layers containing the ink surface coverages of the

classical cmyk inks and the additional daylight fluorescent magenta and yellow inks. These

separation layers are obtained by establishing a relationship between ink surface coverages

and gamut mapped CIELAB values. In a first step, we create a uniform CIELAB grid within the

destination fluorescent gamut and fit the surface coverages . Then, for each input sRGB gamut

mapped color that is to be printed, we apply within the CIELAB grid a 3 dimensional lookup

table base interpolation (Kang 2006) between computed surface coverages.

The uniform 3D CIELAB grid is established by varying CIELAB values L∗, a∗, b∗ by small steps

sp , e.g. sp = 2, from the minimal to the maximal CIELAB values of the destination fluorescent

gamut G f . Some points of the CIELAB grid may not be present within the G f gamut. A point

Cgrid of the uniform CIELAB grid can be removed by creating a line passing through it, a single

point Cb/w in the black and white axis and intersecting this line with the destination gamut

boundary at CdestMax. If the distance from Cb/w to Cgrid is strictly larger than the distance from

Cb/w to CdestMax, this point does not belong to the G f gamut and can therefore be removed

from the uniform CIELAB grid. At the boundary of the destination gamut, in order to prevent

removing one of the 8 necessary CIELAB cube vertices used to perform the 3D interpolation

between mapped CIELAB values and corresponding ink surface coverages, we reduce the

step size sp to 1 and we keep an additional point beyond the destination gamut. For each

CIELAB values of the so established CIELAB grid, we compute the corresponding ink surface

coverages by performing a gradient descent on the IS-CYNSN model, i.e. by minimizing the

∆E94 differences between predicted spectral reflectances converted to CIELAB and CIELAB

grid values. The minimization is done for each ink set. We store in 4 lookup tables the mapping

between CIELAB grid values and cmyk, cm f y f , cm f y and cmy f fitted surface coverages as

well as corresponding ∆E94 prediction differences. The minimizations are carried out with the

fmincon Matlab operator. In addition, in order to ensure the grey component replacement

(GCR), we build the cmyk lookup table by constraining the cmyk separation on the lightness,

as is usual in most cmyk printing systems.

Finally, the ink separation layers are generated by performing a tri-linear interpolation between

mapped input image CIELAB points and fitted surface coverages of the 8 surrounding vertices

of the CIELAB grid. We also perform a tri-linear interpolation in order to obtain interpolated

prediction differences. We then test if the mapped color can be reproduced with one of the ink

set. This is the case when one of the interpolated prediction differences shows a negligible

∆E94 difference. In order to minimize the amount of fluorescent ink, we test the ink sets in the

order cmyk, cm f y , cmy f and cm f y f . Since we have performed the gamut mapping for the

union G f of the gamuts of the ink sets, we ensure that at least one ink set is able to reproduce

the given color.

46

4.7. Summary of application user-defined parameters

4.7 Summary of application user-defined parameters

In order to print gamut expanded images, we developed a prototype software. This software

maps input sRGB images according to the many different user-defined parameters presented

in this chapter. In this section, we summarize and discuss these parameters in the list that

follows:

- Theδ chroma expansion limitation factor limits the maximal effective chroma expansion

achievable by the fluorescent gamut, where 0 ≤ δ ≤ 1. With δ = 1, the limitation is

maximal and no color beyond sRGB gamut colors is printed. In order to take advantage

of the large fluorescent gamut with the goal of having a better reproduction of input

image sRGB colors, we set this parameter to one. With δ = 0, we use the maximal

expansion achievable with the fluorescent gamut G f .

- The γ non-linear chroma reinforcement factor increases non-linearly chroma and

brightness towards high chroma and bright fluorescent colors (0 < γ≤ 1). With γ= 1,

we do not reinforce colors. We decrease its values depending on how much we want to

reinforce colors of image regions of special interest.

- The lh high lightness focal point controls the positive lightness shift when expand-

ing source colors. By decreasing its value, we map input sRGB color towards higher

lightnesses. This parameter depends on the destination gamut boundary shape. For

the considered inkjet fluorescent gamut we set lh = 76 and for the considered offset

fluorescent gamut we set lh = 85.

- The ll low lightness focal point controls the negative lightness shift when reducing

source colors. This parameter is used to avoid reducing too much chroma in dark tones.

This parameter depends on the destination gamut boundary shape. For the considered

inkjet fluorescent gamut we set ll = 40 and for the offset fluorescent gamut we set ll = 30.

- The ε display tri-stimulus reduction factor simulates a display with lower tri-stimulus

phosphor values. The user or the application (in automatic mode) set this parameters

until no mapped color saturates the sRGB display Red, Green and Blue channels. This

parameter enables visualizing the difference between the expanded and non-expanded

image parts of the previewed destination gamut mapped image.

- The κ distance limitation factor creates smooth chroma transitions along the selected

region boundary.

- The P1, P2 lightness adaptation function parameters control the lightness from the input

sRGB space to the selected printable output gamut. Since linear lightness mapping

better preserves lightness differences present in the original input image, we use it for

the results presented in the next chapter.

47


4.8 Summary

We propose a framework for expanding the colors of sRGB images towards printable high

chroma and bright colors located beyond display sRGB gamut colors. High chroma and

bright colors are obtained by performing a gamut expansion of the original image gamut

onto the gamut covered by the combined cmyk and the magenta and yellow fluorescent inks.

Fluorescent ink halftones add a new dimension to color prints. They enable either better

reproduction of sRGB colors or enable highlighting image parts to attract the attention of

the observer. Applications include the design of posters and images for advertisement, as

well as improved reproductions such as watch images or art paintings. The proposed color

reproduction framework enables users to choose (a) the image regions to be enhanced, (b)

how far the chroma should be expanded and (c) the possible non-linearity of the chroma

expansion. User can display a preview of the gamut expanded image print. This is useful for

designers working in fields such as photography, advertisement and production of catalogues

and magazines as well as for artists who want to create new design effects.

48

5 Gamut expanded images

5.1 Introduction

In this chapter, we show gamut expanded images generated with the developed prototype

software. Gamut expansion of these images is controlled according to the gamut expansion

parameters presented in the previous chapter. In a first step, we validates our preview software

by showing a preview of a gamut expanded image together with its corresponding print (Sec-

tion 5.2). Then, we show printed gamut expanded images that illustrate the new possibilities

offered by the fluorescent gamut. These images comprise designs for advertisement (Section

5.3), images of watches and master painting whose colors are better reproduced with the

fluorescent gamut than by using a classical cmyk gamut (Section 5.4) and images of artistic

designs (Section 5.5).

The previewed images where displayed on an calibrated Eizo ColorGraphic CG245W monitor

and corresponding prints are printed either on an Canon MP-101 paper with an inkjet Epson

P50 printer with original cmyk inks and the inkjet daylight fluorescent magenta and yellow

inks or on a HEAVEN 42 SOFTMATT coated paper with an offset 6 ink Heidelberg printer with

classical cmyk inks and the Pentone offset daylight fluorescent magenta and yellow inks. All

figures contain photographs of the prints as well as the previewed prints taken by a Canon

PowerShot S95 camera under normal daylight conditions.

5.2 Preview and corresponding print of a gamut expanded image

In this section, we show an example of gamut expanded images previewed on a display and

printed.

Figure 5.1a shows a photograph of the print preview of a lizard where the selection comprising

the animal and its boundaries has been gamut expanded with a chroma non-linear reinforce-

ment factor γ = 0.3 without smooth chroma transition between the gamut expanded and

non-gamut expanded part. Figure 5.1b shows the same display preview with smooth chroma

49

Chapter 5. Gamut expanded images

transitions along the boundary formed by the gamut expanded and non-gamut expanded

image part. The corresponding spatial interpolation map using a distance limitation factor

κ = 2.5 is shown in Figure 4.6. In addition, in order to render out of sRGB printable gamut

colors, a tri-stimulus display reduction factor of ε= 0.88 has been applied. Figure 5.1c shows a

photograph of the inkjet printed image previewed in Figure 5.1b. Figure 5.1d shows a photo-

graph of the same lizard image printed without chroma expansion with classical cmyk inkjet

inks only.

(a) (b)

(c) (d)

Figure 5.1: Photographs of (a) the display preview of a lizard image where a selection compris-ing the lizard as well as its boundary pixels has been gamut expanded and (b) the same displaypreview where smooth chroma transitions have been created between the inner and outerparts of the selection, (c) the image with smooth chroma transitions printed with fluorescentand classical inks and (d) the classical cmyk print of the lizard image. Please observe theimages on the electronic version of the thesis.

In Figure 5.1a the rectangle regions show along the selection boundaries strong color artefacts

due to the high differences in chroma and in brightness between the gamut expanded and

non-gamut expanded parts of that image. For instance, below the mouth of the animal, some

bright colors have been strongly gamut expanded while neighboring darker colors have not

50

5.3. Advertising images

being gamut expanded. This yields in this rectangular area strong color differences between

neighboring pixels. By creating smooth chroma transitions with a spatial interpolation map,

these color artefacts disappear as it is apparent within the rectangular regions of Figure 5.1b.

By comparing the printed gamut expanded image (Figure 5.1c) with its corresponding display

preview (Figure 5.1b), we observe that the non-expanded part of the image have similar colors

both in the preview and in the print. However, the gamut expanded part within the head of

the lizard have different colors, i.e. in the photograph, the print colors appear brighter and

less saturated. This mainly due to the fact that colors of the print within the selected image

part are beyond sRGB colors and can therefore not be rendered with a sRGB image captured

by a digital camera. Finally, by comparing the printed gamut expanded image with a classical

cmyk print of that image (Figure 5.1d), we observe that within the selected image part (lizard)

colors are brighter and have a higher chroma. In Figure 5.1c, since the G f gamut covers a

larger part of the GsRGB gamut compared with Gcmyk gamut coverage, the image parts outside

the selection where no chroma expansion has been applied are also better reproduced. For

example, in the printed fluorescent ink image, the stone floor colors match the original sRGB

image while the stone colors in the cmyk printed image (Figure 5.1d) do not match the original

sRGB stone colors.

5.3 Advertising images

In this section, we show examples of printed gamut expanded images that could be used for

advertisement. These images have been produced by selecting image parts and by increasing

chroma and brightness of their colors. The goal is to either reinforce the observer attention to

a specific region of the image, for example to attract its attention on the product that is to be

sold or to highlight some image parts.

Figure 5.2 shows photographs of the Rolex Yachtmaster watch advertising image produced

with the considered inkjet printer (a) non-gamut expanded and printed with classical cmyk

inks and (b) the same image with the watch being gamut expanded with a non-linear chroma

reinforcement factor of γ= 0.3. By expanding the chroma of the watch colors, the attention

of the observer will be directed towards the watch. By comparing the man in both prints, we

observe that even without being gamut expanded, the face and hair are better rendered and

have a higher contrast when reproduced with the G f gamut (Figure 5.2b). They better preserve

the original Yachtmaster advertising image.

For the watch shown in Figure 5.2b, Figure 5.3 represents the y , y f , m and m f ink layers in

grayscale. The darkness at each location represents the ink surface coverage. This yields the

brighter and higher chroma colors of the watch shown in Figure 5.2b.

Figures 5.4 shows photographs of a printed Mercedes AMG car both gamut expanded with

the offset fluorescent gamut and non-gamut expanded with the classical cmyk offset gamut.

In order to compare printed gamut expanded images with a standard industrial print of that

image, the color separation layers of the classical cmyk print of that image have been generated

51


(a) (b)

Figure 5.2: Photographs of a printed (a) non-gamut expanded Rolex Yachtmaster image printedaccording to the classical Gcmyk gamut and (b) of the same image printed according to the G f

gamut, where the selected watch is gamut expanded with a non-linear chroma reinforcementfactor γ= 0.3. Please observe these images on the electronic version of the thesis.

(a) (b) (c) (d)

Figure 5.3: The yellow (a), daylight fluorescent yellow (b), magenta (c) and daylight fluorescentmagenta (d) ink layers of the gamut expanded Rolex Yachtmaster watch image. Layers areshown in grayscale with the darkness at each location representing the ink surface coverage.

with the standard ECI ISO Coated V2 profile. In Figures 5.4b and 5.4c, we linearly and non-

linearly expanded colors of the car by using respective chroma reinforcement factors γ= 1 and

γ= 0.5. By comparing the standard print of that image (Figure 5.4a) with its corresponding

linearly expanded print (Figure 5.4b), we observe that colors of the car are more bright and

52

5.3. Advertising images

saturated in the gamut expanded image. This reinforce light reflects in the front parts and in

the doors of the car. These reflects are even more pronounced with the image produced with a

non-linear chroma reinforcement factor (Figure 5.4c). We also observe that the background of

the image is better reproduced with the G f gamut than with the Gcmyk gamut. For instance,

lights of the dome are significantly better reproduced by printing them with combinations of

classical cmyk inks and daylight fluorescent inks (Figures 5.4 b and c) than by printing them

with classical cmyk inks only (Figure 5.4a).

(a) (b)

(c)

Figure 5.4: Photographs of a printed (a) non-gamut expanded AMG car image with the classicaloffset Gcmyk gamut, produced with the standard ECI ISO Coated V2 profile and (b) of the sameimage printed with the offset G f gamut, where a selection comprising the car is linearly gamutexpanded and (c) non-linearly gamut expanded with a non-linear chroma reinforcementfactor γ= 0.5. Please observe these images on the electronic version of the thesis.

The last example shows a lipstick advertisement. Lipstick are used to reinforce colors of the

lips by reflecting strong saturated and bright colors. By using the fluorescent gamut instead of

the classical cmyk gamut, it is possible to better reproduce these colors as well as to create the

desired attractive effect on the lips. Figure 5.5 shows both gamut expanded and non-gamut

expanded offset prints of a home made design of a lipstick advertisement. Figure 5.5a shows

the classical cmyk print of that design using the standard ECI ISO Coated V2 profile and Figure

5.5b shows its corresponding gamut expanded print where a selection comprising the lips of

the girl as well as the lipstick has been linearly gamut expanded. Since by superposing the m f

53


(a) (b)

Figure 5.5: Photographs of a printed (a) non-gamut expanded advertisement lipstick imagewith the classical Gcmyk gamut, produced with the standard ECI ISO Coated V2 profile and (b)of the same image printed with the G f gamut, where a selection comprising the lips as well asthe lipstick is linearly gamut expanded. Please observe these images on the electronic versionof the thesis.

ink with the y f ink we obtain a higher chroma and brighter red color than by superposing the

m ink with y ink (see Chapter 1, Figure 1.3), red colors of the lips and the lipstick are reinforced

in the gamut expanded print. In addition, since the G f gamut is larger than the Gcmyk gamut,

the face of the girl is better reproduced in the gamut expanded image.

5.4 Better reproduction of input sRGB image colors

In this section, we show that input sRGB colors are significantly better reproduced by using the

fluorescent G f gamut than by using a classical cmyk gamut. Better reproduction of input sRGB

image colors is obtained by limiting the maximal expansion achievable with the fluorescent

G f gamut to the sRGB gamut and by not reinforcing input chroma, i.e. by using a maximal

effective expansion factor δ= 1 (no beyond sRGB colors are printed) and a non-linear chroma

reinforcement factor γ= 1 (no chroma is reinforced). Examples show offset printed images

of watches and master paintings. As a comparison, we also show classical industrial cmyk

prints whose color separation layers have been generated with the standard ECI ISO Coated

V2 profile.

The first example shows photographs of offset prints of a pink gold Hublot watch printed with

the ECI ISO Coated V2 profile (Figure 5.6a) and the same image printed with the G f gamut

(Figure 5.6b). Since the G f gamut is larger than the Gcmyk gamut in both the bright magenta

and red regions of the gamut (See Chapter 3), pink gold colors are better reproduced in the

fluorescent print. Appropriate low ink surface coverages of the m f ink (Figure 5.6c) enable

achieving improved printed pink gold colors. In addition, since the prototype software is

calibrated for the specific printer, paper, inks and machine parameters, we are able to achieve

a better black than by using a standard profile that is used for various papers and inks. With

54

5.4. Better reproduction of input sRGB image colors

100%, 73%, 4% and 100% respective cmyk ink surface coverages for the fluorescent print, we

obtain a black at CIELAB L∗ = 15, a∗ = 0.9 and b∗ =−0.8 values while the black obtained with

the ECI ISO Coated V2 profile gives CIELAB values at L∗ = 20, b∗ = 1 and a∗ = 4 with 87%,

78%, 65% and 93% respective cmyk ink surface coverages. By using less printed inks in the

fluorescent print (total surface coverage = 277%) than in the classical cmyk print (total surface

coverage = 323%), we obtain a deeper black that is less colored. Compared to the classical

cmyk print of the Hublot watch, the fluorescent print is of higher contrast and reflects more

vivid colors.

(a) (b) (c)

Figure 5.6: Photographs of a printed (a) advertisement Hublot pink gold watch with theclassical Gcmyk gamut, with color separation layers produced with the standard ECI ISOCoated V2 profile and (b) of the same image printed with the G f gamut. In addition, image (c)shows the m f ink layer of the image (b) where the darkness at each location representing theink surface coverage. Please observe these images on the electronic version of the thesis.

Figure 5.7 shows photographs of offset prints of a Rolex watch (a) printed with the standard

cmyk ECI ISO Coated V2 profile and (b) printed with the fluorescent gamut. By comparing

these two images, we observe that yellow gold colors are better reproduced with the G f gamut.

Gold colors of the Rolex watch printed with the Gcmyk gamut are reproduced with the yellow

ink. Since the offset yellow ink does not reflected enough light in the red visible wavelengths,

the reproduced gold color are greenish. By adding a little amount of m f ink (Figure 5.7c) to

the yellow ink, we increase the reflected light in the red wavelengths, thereby obtaining a color

closer to the yellow gold color. With a deeper black, improved red rubis and gold colors, the

fluorescent print of the Rolex watch has a higher contrast and is more colorful.

Note that these two watch images have been directly reproduced from a watch photograph

without pre-press retouching work. Images printed on catalogues are modified until customer

requirements are achieved. The images shown here can therefore not be compared with the

images printed in official Rolex and Hublot catalogues.

Master painting are known to have colors which are beyond classical cmyk printer gamut

55


boundaries. Impressionist painters, such as for instance Claude Monet, Joseph Mallord

William Turner and Paul Gaugin have produced master paintings with high chroma yellow,

red, magenta and orange colors. These paintings can therefore be better reproduced with a

fluorescent gamut having larger boundaries in these color regions than with a classical cmyk

gamut.

(a) (b) (c) (d)

Figure 5.7: Photographs of a printed (a) advertisement Rolex yellow gold watch printed withthe classical Gcmyk gamut, with color separation layers produced with the standard ECI ISOCoated V2 profile and (b) of the same image printed with the G f gamut. In addition, image (c)and (d) shows the respective m f and y f ink layers of the image (b) with the darkness at eachlocation representing the ink surface coverage. Please observe these images on the electronicversion of the thesis.

Figure 5.8 shows photographs of the master painting "San Giorgio Maggiore" by Claude Monet

printed according to (a) the offset classical Gcmyk and (b) the offset fluorescent G f gamut. By

comparing these two prints, we observe that strongly saturated red, orange, yellow and blue

colors of this master painting are significantly better reproduced in the fluorescent print.

J. M. W. Turner is an english artist who painted sunset and sunrise landscapes. In "The Fighting

Temeraire", the half right part of this master painting shows a chip in a esturary at sun set. Sun

light goes through the clouds and the flaming red of the clouds is reflected by the river. These

flaming colors have been better reproduced with the offset fluoresent gamut G f (Figure 5.9b)

than with the classical cmyk gamut (Figure 5.9a). In addition, since the G f gamut is larger

than the classical cmyk gamut, we observe more details in the fluorescent print of this master

painting. For instance, we observe more details in the water as well as in the painted blue

regions.

56

5.4. Better reproduction of input sRGB image colors

(a) (b)

Figure 5.8: Photographs of the master painting "San Giorgio Maggiore" by Claude Monet (a)printed with the classical offset Gcmyk gamut, according to the standard ECI ISO Coated V2profile and (b) of the same image printed with the offset G f gamut. Please observe theseimages on the electronic version of the thesis.

(a) (b)

Figure 5.9: Photographs of a part of the master painting "The Fighting Temeraire" by J. M. W.Turner (a) printed with the classical offset Gcmyk gamut, according to the standard ECI ISOCoated V2 profile and (b) of the same image printed with the offset G f gamut. Please observethese images on the electronic version of the thesis.

The last example of this section shows offset prints of the master painting "Women of An-

davadoaka" by Paul Gaugin. By comparing the classical cmyk print of that master painting

(Figure 5.10a) with its corresponding fluorescent print (Figure 5.10b), we observe that the

fluorescent gamut better renders brown skin colors as well as yellow and orange background

colors. We also observe that green colors of the trees and the clothes of one of the painted girl

are better rendered in the fluorescent print.

57


(a) (b)

Figure 5.10: Photographs of the master painting "Women of Andavadoaka" by Paul Gauguin(a) printed with the classical offset Gcmyk gamut, according to the standard ECI ISO CoatedV2 profile and (b) of the same image printed with the offset G f gamut. Please observe theseimages on the electronic version of the thesis.

5.5 Artistic images

In this section, we show examples of artistic prints obtained by increasing chroma and bright-

ness of input sRGB image colors. The resulting printed images have vivid colors beyond sRGB

gamut colors.

Figure 5.11 shows a flower image (a) printed with the offset Gcmyk gamut, (b) printed with the

offset G f gamut and non-gamut expanded with a chroma reinforcement factor γ= 1 and a

chroma expansion limitation factor δ= 1 (no beyond sRGB colors are printed), and (c) printed

with the offset G f gamut and non-linearly gamut expanded with a chroma reinforcement factor

γ= 0.8. By comparing the non-fluorescent print with the non-gamut expanded fluorescent

print, we observe that colors of the input sRGB image are better reproduced in the fluorescent

print. By applying to the input image a non-linear chroma reinforcement factor (Figure 5.11c),

we rapidly increase the chroma of input sRGB colors towards strongly bright and high chroma

daylight fluorescent colors.

The next example shows printed reproductions of an image specially designed for offset

fluorescent prints. A graphical designer used our prototype proofing software to design a

magenta flaming girl image with the goal of obtaining interesting printed fluorescent colors.

Figure 6.7a shows the classical cmyk offset print of that image and Figure 6.7b shows the

linearly gamut expanded fluorescent print of that image. By using combinations of classical

inks with daylight fluorescent inks, we are able to create a strong flaming effect around the

girl. Flames in the fluorescent print appear to be strongly yellow and of high chroma magenta

colors. In the classical cmyk print, flames appear to be bright and of a lower chroma. In the

fluorescent print, both the background black color is deeper and colors have a higher chroma.

58

5.5. Artistic images

(a) (b) (c)

Figure 5.11: Photographs a of flower image (a) printed with the classical offset Gcmyk gamut,according to the standard ECI ISO Coated V2 profile, (b) of the same image non-gamutexpanded and printed with the offset G f gamut, and (c) non-linearly gamut expanded andprinted with the fluorescent offset G f gamut. Please observe these images on the electronicversion of the thesis.

These intense colors cannot be obtained with classical cmyk inks.

Figure 5.13 shows of a fluorescent mushroom (a) printed with the classical inkjet Gcmyk

gamut and (b) linearly gamut expanded and printed with the inkjet fluorescent G f gamut.

Superposing the inkjet cyan and daylight fluorescent yellow ink yields the high chroma and

bright green colors, enabling creating the fluorescent effect of the mushroom. This is effect is

not achievable with classical inkjet cyan and yellow inks, as shown in Figure 5.13a.

(a) (b)

Figure 5.12: Photographs of a designed flaming girl (a) printed with the classical offset Gcmyk

gamut, according to the standard ECI ISO Coated V2 profile and (b) of the same image linearlygamut expanded and printed with the fluorescent offset G f gamut. Please observe theseimages on the electronic version of the thesis.

59


(a) (b)

Figure 5.13: Photographs of a fluorescent mushroom image (a) printed with the classicalinkjet Gcmyk gamut and (b) of the same image linearly gamut expanded and printed with thefluorescent inkjet G f gamut. Please observe these images on the electronic version of thethesis.

5.6 Summary

The printed gamut expanded images shown in this chapter illustrate the new possibilities

offered by establishing a 6 ink print setup combining the classical cmyk inks complemented

with the daylight fluorescent magenta and yellow inks. These fluorescent inks add a new

dimension to classical cmyk prints. They enable printing image regions with high chroma

and bright colors. This can be used for advertising by highlighting within the destination

fluorescent image the product that is to be sold or by reinforcing the attraction of the observer

on a specific product, e.g. the lips and the lipstick of the lipstick advertisement image. Since

the fluorescent gamut is significantly larger than the classical cmyk gamut, it also enables

better reproducing input sRGB image colors. We showed that compared to classical cmyk

prints, fluorescent prints significantly improved the printer reproduction capabilities for

watches and also for master paintings which are known to have color outside the gamut of

classical inks.

The developed prototype software enables controlling the gamut expansion and displays a

preview of the printable destination gamut expanded image. A graphic designer showed that

is is possible to design images conceived to make use of the fluorescent gamut. He created

both an advertising lipstick image and a flaming girl artistic image with interesting vivid colors

that cannot be obtained with classical cmyk printers.

60

6 Hiding patterns with daylight fluores-cent inks

6.1 Introduction

In this chapter, we propose a method for hiding security patterns within printed images by

making use of classical and of the two daylight fluorescent magenta and yellow inks. Under the

D65 illuminant, we establish in the CIELAB space the gamut of a classical cmyk printer and the

gamut of the same printer using a combination of classical inks with daylight fluorescent inks.

For the two categories of previously considered offset and inkjet printers, these gamuts show

that a significant part of the classical ink gamut can be reproduced by combining classical

inks with daylight fluorescent inks.

Parts of images are either printed with classical inks (with the ink set cmyk) or printed with

combinations of classical inks with one or two daylight fluorescent inks (ink sets cm f y f , cm f y ,

cmy f ). By applying a metameric color match under the D65 illuminant between the ink set

comprising no daylight fluorescent ink and the ink sets comprising daylight fluorescent inks,

we create images which look the same under normal daylight. By changing the illumination,

for example by observing the image under a tungsten, a colored blue or a UV illumination, we

reveal the security patterns formed by the parts of the image printed with daylight fluorescent

inks (Section 6.2). By spatially interpolating between the parts of the image printed with a

fluorescent ink set, e.g the cmy f ink set and the parts of the image printed with the non-

fluorescent ink set cmyk, we are also able to hide a variable intensity (grayscale) image within

a printed full color image (Section 6.3).

In the last section of this chapter, we also propose a method for hiding security patterns

at the same time under the most common illuminations, such as for instance a tungsten,

a fluorescent tube F7, and a daylight illumination. These patterns are revealed under an

illumination having energy only in the excitation wavelength range of the daylight fluorescent

inks (Section 6.4). They can be typically revealed under a colored blue or a UV illumination.

Security features relying on fluorescent inks are not new. For instance, single invisible fluores-

cent inks are widely used in passports, bank notes and credit cards (Van Renessse 2005). The

61

Chapter 6. Hiding patterns with daylight fluorescent inks

hidden patterns are generally printed with a single invisible fluorescent ink, such as the yellow

"VISA" text appearing under a UV light source. Other security features relying on fluorescent

inks are described in Chapter 1 under the section reviewing the prior art (Section 1.6). Daylight

fluorescent inks offer a better protection compared with invisible fluorescent inks. In order

to print colors of the original image there is a need of establishing a gamut mapping from

the original image color space, e.g. the sRGB display gamut to the ink destination gamut. In

addition, there is also a need of finding an exact relationship between gamut mapped original

image colors and corresponding fluorescent and non-fluorescent ink dot surface coverages.

This can be achieved only with a spectral prediction model dedicated for predicting spectral

reflectances of halftones comprising daylight fluorescent inks. Finally, hidden patterns printed

with invisible fluorescent inks are revealed only under a UV illumination. Hidden patterns

printed with daylight fluorescent inks are not only revealed under a UV illumination but also

under a visible illumination different from the reference illumination such as a filtered blue

illumination.

Finally, this method for hiding security patterns is fully compatible with the gamut expanded

image framework presented in Chapter 4. Once a print company has integrated this framework

with its printing workflow, this company can propose to its customers this security feature

without additional effort. Compared to security features relying on invisible fluorescent inks,

this security feature does not need any invisible security UV inks whose diffusion is restricted

to companies active in the field of security. The protection of the proposed security feature

with daylight fluorescent inks is embedded into the way of how these fluorescent inks are

printed and not into the fluorescent emission characteristic of the fluorescent inks.

6.2 Hiding security patterns by printing colors either with or with-

out daylight fluorescent inks

In Chapter 1 Section 1.3 we have shown that is possible to create interesting colorants by

superposing classical inks with daylight fluorescent inks or by superposing several daylight

fluorescent inks. With these fluorescent colorants, we establish a strictly fluorescent gamut

Gs f that comprises all colorants combining classical inks with a least one daylight fluorescent

inks. More precisely, the strictly fluorescent gamut Gs f is the conjunction of the three cm f y f ,

cm f y , cmy f ink set fluorescent sub-gamuts, i.e. one color of the strictly fluorescent gamut Gs f

is associated with at least one of these sub-gamuts and is printed with its corresponding 3 inks.

By comparing the Gs f gamut with the classical ink Gcmyk gamut under the D65 illuminant, we

determine the colors of the Gcmyk gamut which are metameric to the colors of the Gs f gamut

under normal daylight conditions. We print the hidden patterns whose colors are located

within the Gs f gamut with a fluorescent ink set so as to have an exact metameric match with

the gamut mapped original color printed with the classical ink Gcmyk gamut viewed under the

D65 illuminant. These patterns will therefore be hidden under normal daylight.

Figure 6.1 illustrates in the CIELAB space a comparison between the inkjet Gcmyk and Gs f

62

6.2. Hiding security patterns by printing colors either with or without daylightfluorescent inks

a*-50 0 50

-40

0

40

80b*

L* = 45

Gcmyk

Gsf

0

-40

40

80

b*

L* = 60

Gcmyk

Gsf

a*-50 0 50

L* = 75

GcmykGsf

a*-50 0 50

L* = 55

L* = 70

Gcmyk

Gcmyk

Gsf

Gsf

-40

0

40

80

b*

-40

40

80

b*

0

a* 50-50 0

a*-50 0 50

-40

0

40

80

b*

L* = 85

GcmykGsf

-40

0

40

80

b*

a* 50-50 0

L* = 50

L* = 65

Gcmyk

Gcmyk

Gsf

Gsf

-40

0

40

80

b*

-40

40

80

b*

0

a* 50-50 0

L* = 80

GcmykGsf

-40

0

40

80

b*

a*-50 0 50a*-50 0 50

inkjet

Figure 6.1: Color gamut Gcmyk of the classical cmyk inkjet ink set (solid lines) and the strictlyfluorescent inkjet gamut Gs f (dotted lines).

gamut boundaries under the D65 illuminant. For lightnesses less than L∗ = 55, we observe

that the Gs f gamut is smaller than the Gcmyk gamut. This can be explained by the fact that

the daylight fluorescent colorants are brighter than the corresponding classical colorants

(see Chapter 1, Figure 1.2). At a lightness between L∗ = 55 and L∗ = 65, there are not many

differences between the classical ink gamut (Gcmyk ) and the restrictive fluorescent gamut

(Gs f ), i.e. only a small part of the Gcmyk gamut is outside of the Gs f gamut. For lightnesses

higher than L∗ = 65, the Gcmyk gamut is included within the G f gamut. We therefore observe

that for bright CIELAB colors (having a lightness L∗ > 55), we are able to reproduce most of

the classical inkjet cmyk colors by combining classical and daylight fluorescent inks.

Figure 6.2 illustrates in the CIELAB space a comparison between the offset Gcmyk and Gs f

gamut boundaries under the D65 illuminant. For lightnesses less than L∗ = 35, we observe

that the Gs f gamut is smaller than the Gcmyk gamut. For lightnesses between L∗ = 35 and

L∗ = 55, only a small region of the Gcmyk gamut has larger boundaries than the Gs f gamut.

For lightnesses higher than L∗ = 55, the Gcmyk gamut is strictly included within the Gs f gamut.

In this situation, for mid-bright colors having a lightness higher than L∗ = 35 and less than

63


Gcmyk

Gsf

b*b*

-50

0

50

L* = 35

-50 0 50a* -50 0 50

a*

b*

-50

0

50

L* = 45

b*

-50

0

50

-50 0 50a*

L* = 55

b*

-50

0

50

-50 0 50a*

L* = 60b*

-50

0

50

-50 0 50a*

L* = 65

b*

-50

0

50

a*-50 0 50

L* = 70

b*

-50

0

50

-50 0 50a*

L* = 75

b*

-50

0

50

-50 0 50a*

L* = 80

-50 0 50a*

b*

-50

0

50

L* = 85

GcmykGcmyk

GcmykGcmyk

Gcmyk

GcmykGcmykGcmyk

GsfGsf

GsfGsfGsf

GsfGsf

Gsf

offset

Figure 6.2: Color gamut Gcmyk of the classical cmyk offset ink set (solid lines) and the strictlyfluorescent offset gamut Gs f (dotted lines).

L∗ = 55, we are able to reproduce most of the classical offset cmyk colors by combining offset

classical and daylight fluorescent inks. For bright colors having a lightness higher than L∗ = 55,

we are able to reproduce all offset cmyk colors by combining classical offset cmyk inks with at

least one daylight fluorescent ink.

For hiding security patterns within an image, we define a mask. The mask can represent any

patterns such as for instance the security "VALID" text shown in Figure 6.3. While generating

a specific image, we print outside the mask the colors of the image with the Gcmyk gamut,

i.e. with classical inks only. Inside the mask, if colors of the Gcmyk gamut are reproducible

by colors of the Gs f gamut, we print them with fluorescent colorants, i.e. combinations of

classical and daylight fluorescent inks. In the contrary case, we use classical inks only.

The next challenge consists in establishing an exact relationship between the CIELAB colors

and the ink surface coverages of the inks defining either the Gcmyk gamut or the Gs f gamut.

This relationship must be exact in order to print perfectly metameric colors. This exact rela-

tionship is established thanks to the IS-CYNSN spectral prediction model that is remarkably

64

6.2. Hiding security patterns by printing colors either with or without daylightfluorescent inks

VALID Gsf or Gcmyk

Gcmyk

Figure 6.3: Example of an image design incorporating the hidden "VALID" security text.Outside the "VALID" mask, colors are printed with classical cmyk inks only (with Gcmyk ).Inside the "VALID" mask, colors are printed either with classical cmyk inks only (with Gcmyk )or with combinations of classical cmyk and daylight fluorescent inks (with Gs f ).

accurate for predicting total spectral reflectances of halftones comprising classical inks only

or combining classical and daylight fluorescent inks (See Chapter 2, Sections 2.5 and 2.6).

In order to obtain a relationship between the ink surface coverages and the sRGB values of

the image that is to be reproduced, we map all the sRGB CIELAB values by steps of 3% R, G

and B into the Gcmyk gamut. This is achieved by the standard multiple foci gamut mapping

approach as it is explained in Chapter 4. We obtain the ink surface coverages corresponding to

the color mapped into the printer non-fluorescent Gcmyk gamut with the IS-CYNSN model

by minimizing the ∆E94 differences between the predicted color and the desired color. The

minimization is carried out for each ink set. We store the fitted ink surface coverages plus the

corresponding ∆E94 differences between desired and predicted colors. This yields four lookup

tables mapping sRGB values to cmyk, cm f y f , cm f y and cmy f ink surface coverages with

corresponding ∆E94 differences. The minimizations are carried out with the Matlab fmincon

operator. Note that the IS-CYNSN models are calibrated for the D65 illuminant by measuring

the calibration patches with that illuminant.

For generating an image incorporating a hidden pattern we test if the mapped sRGB colors

within the mask can be reproduced by one of the fluorescent ink sets. This is the case when

the corresponding entry in one of the cm f y f , cm f y and cmy f lookup table shows a negligible

∆E94 difference between desired gamut mapped color and the color predicted with the fitted

ink surface coverages. In order to maximize the amount of fluorescent ink, we test the ink

set in the order cm f y f , cm f y , cmy f . If no fluorescent ink set provides the desired color, it is

printed with the classical cmyk ink set. Gamut mapped colors outside the mask are printed

with the classical cmyk ink set.

6.2.1 Illustrations of hidden security patterns

The printed images shown in this section embed the repetitive hidden "VALID" text pattern.

Lookup tables mapping the sRGB values to the ink surface coverages have been generated

for the D65 illuminant. Thus, these patterns are hidden under a normal daylight viewing

65


condition but revealed under both the A or the UV illuminations. Images were printed with

the EPSON P50 printer with native EPSON cmyk inks and with the considered inkjet daylight

fluorescent magenta and yellow inks. Pictures of the prints have been taken with a Canon

PowerShot S95 camera under normal daylight conditions, under UV-A black light and under a

tungsten lamp (A illuminant).

Figure 6.4 illustrates a printed Japanese girl image embedding the repetitive "VALID" hidden

pattern, photographed both under normal daylight (left image) and under UV light (right

image). Under normal daylight conditions it not possible to distinguish the text "VALID"

formed by combinations of classical and daylight fluorescent inks. This is due to the fact that

we have a perfect metameric match between the inner and outer part of the "VALID" mask.

Under UV illumination, the text "VALID" is visible in almost all parts of the image, except

in the hair. Since the hair is dark, it is not possible to reproduce it with daylight fluorescent

colorants.

Figure 6.5 illustrates a printed Iceland landscape embedding the repetitive "VALID" hidden

pattern. While under normal daylight it is not possible to distinguish the hidden pattern,

under both A and UV illuminations, it is revealed. Since the A illuminant has less energy

than the D65 illuminant in the excitation range of the daylight fluorescent inks, there is less

fluorescent emission and therefore the "VALID" mask content appears darker than when seen

under the D65 illuminant, see Figure 6.5c.

(a) (b)

Figure 6.4: Photographs of a printed Japanese girl image incorporating the repetitive "VALID"pattern, (a) viewed under normal daylight and (b) viewed under UV illumination.

6.3 Hiding a variable intensity security image

In this section, we show how to embed a hidden variable intensity image within a printed im-

age. The hidden variable intensity image comprises halftones enabling to spatially interpolate

between a fluorescent ink set and the non-fluorescent classical cmyk ink set. When seen under

66

6.3. Hiding a variable intensity security image

(a) (b) (c)

Figure 6.5: Photographs of a printed Iceland landscape incorporating the repetitive "VALID"pattern viewed under (a) normal daylight, (b) under UV illumination and (c) under A illumina-tion. Please observe the images in the electronic version of the thesis.

a UV illumination, parts printed with a fluorescent ink emit a colored light whose intensity is

proportional to the amount of printed fluorescent ink, i.e. proportional to the intensity level

of the hidden image that is revealed.

In order to hide a variable intensity image within a printed image, we first locate a spatial

region of the gamut mapped destination image whose colors are located at the intersection of

the strictly fluorescent gamut Gs f and the classical cmyk gamut Gcmyk . Intersected Gcmyk and

Gs f gamut color regions are shown for the considered inkjet and offset gamuts respectively in

Figures 6.1 and 6.2. Such a region can for instance be found in the gamut mapped destination

image at locations where colors are bright, i.e. have a lightness L∗ > 55. The variable intensity

image is halftoned with a small diagonally oriented cluster-dot screen (Haines et al. 2003a) or

with stochastic dots generated with a blue noise dither matrix or by error-diffusion (Haines

et al. 2003b) thereby obtaining an image made of black and white pixels. By printing the white

pixels with the fluorescent ink set and the black pixels with the non-fluorescent ink set, we

hide the variable intensity image within the printed image.

By observing the printed image incorporating the hidden variable intensity image under a UV

illumination, parts printed with the fluorescent ink set emit a colored light depending of the

fluorescent ink set used to represent the variable intensity image. For the cmy f ink set, the

fluorescent emission yields greenish colors while the fluorescent ink set cm f y yields reddish

colors. The intensity of the fluorescent emission is proportional to the surface coverage of

printed fluorescent ink. Since these surface coverages have been obtained by halftoning, the

intensity of the colored fluorescent emission is proportional to the intensity of the hidden

variable intensity image. This reveals the hidden variable intensity image.

Figure 6.7 shows an inkjet print of a girl embedding a hidden variable intensity tiger image.

The tiger image is halftoned with blue noise dithering (Figure 6.6). Since the left upper part

of the destination girl image has bright colors, i.e. a lightness L∗ > 55, we choose this image

region to place the hidden tiger. In the destination image we use the halftoned tiger as spatial

67


Gcmyk

Gcmyf

Figure 6.6: Variable intensity tiger image halftoned with a blue noise dithering. This halftonedimage is incorporated within another security image by printing its white pixels with the cmy f

ink set and its black pixels with the cmyk ink set.

mask. As is shown in an enlargement of the mask content (Figure 6.6, red rectangle), we print

destination image colors with the ink set cmy f where the halftoned tiger has white pixels. The

other pixels of the destination image are printed with the ink set cmyk. Since in the destination

image region colors are bright, we are able to reproduce all the mapped colors either with

classical inks only or with combinations of cyan, magenta and daylight fluorescent yellow

inks. While under normal daylight conditions, it is not possible to distinguish the hidden

tiger (Figure 6.7a), under a UV illumination the variable intensity tiger image is revealed by a

colored greenish fluorescent emission of the parts printed with the fluorescent cmy f ink set

(Figure 6.7b).

(a) (b)

Figure 6.7: Photographs of a printed grayscale girl image incorporating a variable intensitytiger image viewed under (a) normal daylight and (b) under UV illumination. Please observethe images in the electronic version of the thesis.

68

6.4. Hiding security patterns under multiple illuminations

6.4 Hiding security patterns under multiple illuminations

In this section, we propose a method for hiding patterns under different illuminations and

revealing them under UV excitation light or under blue light. We would like to hide the patterns

when they are observed under daylight, in an office under a fluorescent tube illumination and

in a room illuminated with tungsten lamps. These patterns should be revealed only under

a narrow band illumination in the excitation wavelength range of the daylight fluorescent

inks, e.g. a UV illumination or a colored blue illumination. We therefore need to optimize the

respective amounts of fluorescent and non-fluorescent inks, so as to produce metamers under

the A and D65 illuminants.

The problem of creating near metamers, called paramers, has been tackled by Urban and

Berns (2011). They proposed a gamut mapping framework that creates mapped colors which

remain substantially similar when observed under different illuminants. Our problem is

different, since by reducing the relative amount of fluorescent ink, the resulting color comes

closer to the color produced by the non-fluorescent cmyk inks, which is the reference color.

In order to create with mixtures of fluorescent halftones and non-fluorescent halftones colors

which are close to the reference cmyk colors under the different illuminants, we proceed as

follows. In a first step, we create a mapping between desired colors and surface coverages

of the fluorescent ink set by minimizing the error between gamut mapped image colors and

predicted colors at the same time for the A and D65 illuminants. Since the color differences are

large when colors are either viewed under the D65 or the A illuminant, we reduce the amount

of printed fluorescent ink within the pattern mask by reducing the ratio of the fluorescent ink

set in respect to the non-fluorescent ink set. This is performed by spatially distributing the

fluorescent ink set by a dither function, which converts relative amounts to surface coverages.

Practical experiments shows that patterns are hidden under various illuminations when no

more than 20% of a daylight fluorescent ink set is used within the pattern mask.

For hiding patterns under a wide range of illuminants, we first select the A and the D65

illuminants. These illuminants have respectively a very low (A) and a very high (D65) energy in

the excitation wavelength range of the daylight fluorescent inks (Chapter 1, Section 1.3). Other

natural illuminations, e.g. cloudy daylight as well as artificial light, e.g. F7 and F11 fluorescent

tubes, provide within the excitation wavelength range of the daylight fluorescent inks energy

that is higher than the energy provided by the A illuminant and lower than the energy provided

by the D65 illuminant. Therefore, patterns hidden under these two illuminants will be hidden

under various natural and artificial illuminations.

For hiding patterns under both the D65 and the A illuminants, while establishing a relationship

between destination image gamut mapped colors and corresponding ink surface coverages,

we not only minimize the ∆E94 difference between gamut mapped color and predicted color

for the D65 illuminant but also at the same time this difference for the A illuminant. We return

as minimization metric the largest minimized∆E94 difference obtained for the two considered

illuminants. Assuming that the minimized difference obtained for the D65 illuminant is given

69


by ∆E D6594 and the minimized difference obtained for the A illuminant is given by ∆E A

94, we

return as fitting metric the maximal minimized ∆E94 error, i.e. we return Max{∆E D65

94 ,∆E A94

}.

In a similar way as presented in Section 6.2, we store for each ink set this maximal minimized

∆E94 error under the D65 and A illuminants plus the corresponding ink surface coverages. For

generating an image incorporating the pattern hidden under different illuminations we test if

the sRGB colors mapped to the classical Gcmyk gamut colors within the pattern mask can be

reproduced by one of the fluorescent ink sets. However, for most gamut mapped sRGB colors,

∆E94 differences are large when colors are viewed either under the A or the D65 illuminant.

We therefore choose the ink set by looking at corresponding entry of the cm f y f , cm f y , cmy f

lookup tables and select the ink set with the smallest joint D65 and A illuminant ∆E94 error. In

a further step, in order to avoid seeing large color differences induced by the large tolerance

on ∆E94 errors, we reduce the amount of daylight fluorescent inks within the mask.

For example, in order to hide the "L" character with combinations of the cyan, magenta

and daylight fluorescent yellow inks, we can create a "L" pattern mask at 20% gray level

intensity. This pattern mask is halftoned with blue noise dithering (Figure 6.8, red rectangle).

We then reduce the amount of printed colors with the cmy f ink set by spatially distributing

this fluorecent ink set according to the halftoned mask pattern. By halftoning the pattern

mask at the same resolution as the resolution of the input image, we avoid to print big clusters

of neighbouring source image pixels with only the fluorescent ink set, i.e. we print isolated

image pixels with the fluorescent ink set cmy f surrounded by many image pixels with the

non-fluorescent ink set cmyk (see Figure 6.8, distributions of the cmyk and cmy f ink sets).

L Gcmyk

Gcmyf

Figure 6.8: Character L of the "VALID" mask message to be hidden under different illuminantsat 20% gray level intensity together with an enlargement of a small region of its correspondingblue noise halftone (red rectangle).

Figure 6.9 illustrates an offset printed Iceland landscape incorporating the repetitive "VALID"

pattern hidden under different white light illuminations. "VALID" patterns are printed with

the ink set cmyk or with the daylight fluorescent ink set cmy f with maximal surface coverage

of the daylight fluorescent yellow ink set at 20%. In this configuration, within the "VALID"

mask, we obtain for the D65 and A illuminants a mean point ∆E94 prediction error of 4.018

70

6.4. Hiding security patterns under multiple illuminations

and a quantile 95% prediction error of 5.24. Since the amount of daylight fluorescent ink is

reduced to 20% surface coverage, it is not possible to distinguish the "VALID" mask content

under normal daylight (Figure 6.9a), under a tungsten illumination (Figure 6.9b) and under

a F7 fluorescent tube illumination (Figure 6.9c). By illuminating the offset printed Iceland

landscape with a blue non UV low consumption lamp, the part printed with the daylight

fluorescent yellow ink is excited by the blue illuminant and emits light that reveals the "VALID"

content. The relative spectral power distribution of the F7 fluorescent tube illumination is

shown in Chapter 1, Figure 1.1.

(a) (b) (c) (d)

Figure 6.9: Photographs of an offset printed Iceland landscape incorporating the repetitive"VALID" pattern viewed under (a) normal daylight, (b) under illumination A, (c) under thefluorescent tube illumination F7 and (d) under a blue low consumption Swiss light classic 55non UV lamp. Please observe the images in the electronic version of the thesis.

Figures 6.10a and b illustrate an inkjet printed Iceland landscape incorporating the repetitive

"VALID" pattern viewed under normal daylight. Valid mask content is printed either with the

inkjet cmyk ink set or with the cmy f ink set for maximal printed daylight fluorescent cmy f

ink set at respectively 10% and at 30%. While it is not possible to distinguish the "VALID"

mask content under normal daylight with maximal printed fluorescent cmy f ink set at 10%

(Figure 6.10a), the "VALID" mask content starts to be slightly visible at 30%(Figure 6.10b, red

rectangle). Figure 6.10c shows the same printed image as shown in Figure 6.10a but viewed

under a UV illumination. Since only a small amount of 10% of the daylight fluorescent ink

set is printed, it is difficult to reveal the "VALID" pattern with a UV illumination. With an

amount of 20% of the printed fluorescent ink set, we obtain an excellent tradeoff between

the capability to be hidden under different illuminations (Figure 6.9) and the capability to be

revealed when seen under a UV illumination (Figure 6.10d).

71


(a) (b) (c) (d)

Figure 6.10: Photographs of an inkjet printed Iceland landscape incorporating the repetitive"VALID" pattern viewed under normal daylight with maximal daylight fluorescent cmy f inksurface coverage at (a) 10%, (b) 30%, (c) photograph of image (a) viewed under UV and (d)photograph of the image with maximal surface coverage of fluorescent cmy f ink set at 20%and viewed under UV illumination. Please observe these images in the electronic version ofthe thesis.

6.5 Summary

We propose a method for hiding security patterns within images by making use of the two

daylight fluorescent magenta and yellow inks. The patterns are printed with combinations of

these two daylight fluorescent inks and classical inks while the rest of the image is printed with

classical inks only. Since the ink surface coverages are calculated with a highly accurate spectral

prediction model calibrated under the D65 illuminant, the embedded security patterns are

completely hidden under normal daylight. By spatially interpolating between a fluorescent ink

set and the classical cmyk non-fluorescent ink set, we are also able to hide a security variable

intensity image within a printed image.

The verification is performed by putting the security images under a tungsten lamp or under a

UV black light and by visually verifying that the security patterns are revealed. With classical

inks it is not possible to hide patterns that are revealed both under UV and A illuminations.

Therefore, these security images are difficult to reproduce.

We also propose a method for hiding the security patterns under different illuminations. By

minimizing the ∆E94 error between gamut mapped colors and predicted colors under at

the same time the A and D65 illuminants and by reducing the part printed with a daylight

fluorescent ink set to a maximal surface coverage at 20%, we showed that the patterns are

hidden under several classical illuminations. The verification is then performed by putting

the security images under an illumination having energy only in the excitation wavelength

range of the daylight fluorescent inks, e.g. a non UV blue lamp or a UV black lamp.

72

7 Conclusion

In this thesis, we explored the possibilities offered by adding the daylight fluorescent magenta

and yellow inks to classical cmyk prints. We propose a complete framework for printing with

daylight fluorescent inks. The solution comprises a spectral prediction model (IS-CYNSN)

that has been optimized to predict spectral reflectances of halftones comprising daylight

fluorescent inks. With a few calibration patches, we achieve remarkable prediction accuracies

for both inkjet and offset halftones printed with classical inks only or with combinations of

classical inks and daylight fluorescent inks (Chapter 2).

Thanks to the accurate spectral prediction of the IS-CYNSN spectral prediction model, we

establish the fluorescent gamut G f comprising colors generated by combining classical inks

with daylight fluorescent inks and the classical cmyk gamut combining classical cmyk inks

only. By comparing these two gamuts for both an inkjet and offset printer, we show that we

considerably expand the domain of printable colors by adding two fluorescent inks to classical

cmyk inks. The domain extension of colors is mainly available for green, yellow, red, magenta

and orange colors (Chapter 3).

These fluorescent inks enable to print high chroma and bright colors that cannot be obtained

with classical inks only. They add therefore a new dimension to color prints. In order to fully

utilize these high chroma and bright fluorescent colors, we define an adapted gamut mapping

from the sRGB display gamut to the fluorescent gamut which allows by gamut expansion to

print beyond sRGB gamut colors. The gamut expansion experience can be user-driven with a

prototype software. This software enables user (a) choosing image regions to be enhanced, (b)

how far and how fast the chroma should be expanded, (c) creating smooth chroma transitions

along the boundary between the gamut expanded and non-gamut expanded image part and

(d) displaying a preview of the printable gamut expanded image (Chapter 4).

This prototype software is useful for designers working in fields such as photography, ad-

vertisement, production of magazine as well as for artists. With this prototype software we

designed and printed images such as a Rolex and a lipstick advertising image where colors of

the products, respectively the watch, the lips and the lipstick are reinforced in order to attract

73

Chapter 7. Conclusion

the attention of the observer. We also show that without performing color gamut expansion,

the large fluorescent gamut enables better reproducing colors of input sRGB images. We

compared offset classical cmyk prints with offset fluorescent prints of watches and master

painting images. The reproduction was significantly improved thanks to the fluorescent inks.

In addition, a graphic designer used our prototype software in order to design and print images

with interesting vivid fluorescent colors (Chapter 5).

In Chapter 6, we show that with these daylight fluorescent inks it is possible to provide

strong optical document security features. Hidden image patterns are printed with daylight

fluorescent inks so as to obtain a perfect metameric match with the colors produced by cmyk

inks only. Parts of image printed with daylight fluorescent inks define a security pattern or a

security variable intensity image that can be revealed under an illuminant having an energy

different from the reference illuminant in the excitation wavelengths of the fluorescent inks.

The process of hiding and revealing the security patterns can be performed with a "hidding"

and a "revealing" illuminant pair. We can for instance hide the security patterns under

normal daylight and reveal them under a tungsten illumination or hide the security patterns

under a tungsten illumination and reveal them under a fluorescent tube F7 illumination. It

is also possible to hide the security patterns at the same time under different natural and

artificial white illuminants such as a normal daylight, a tungsten lamp and a fluorescent

tube illumination. In this situation, the security patterns are revealed under an illumination

having energy only in the excitation wavelengths of the fluorescent inks, such as a blue or a

UV illumination. These security features are easily integrable into the workflow of a printing

company. Once a printing company has integrated in its printing workflow the additional

two fluorescent ink printing stages, it can offer to its customer both higher fidelity printing

capabilities and means of hiding security patterns.

Future work

Additional research is needed to determine user preferences, e.g. which lightness adaptation

strategy is preferable for different types of color images and to which extent the printable color

domain extension offered by the different inks is really perceived and appreciated by the users.

Visual preference experiments would be interesting to investigate the effectiveness of the

proposed gamut expansion algorithm compared to the results obtained by other algorithms

mostly proposed to exploit the gamut of modern wide gamut displays.

Future work should also includes studies about the extent to which chroma expansion of

prints improves the communication of its embedded message.

Regarding the proposed security feature, the security of the hidden patterns can be further

enhanced by establishing a model predicting the fluorescent emission of the daylight fluo-

rescent inks under UV light. By comparing the image captured under a UV illuminant and

the predicted fluorescent image, one may obtain a further confirmation of the authenticity

of the document. Future work should finally verify if the metameric index can be used for

74

expressing the pattern hiding capabilities of different substrates and daylight fluorescent inks

under different daylight illuminants.

75

Bibliography

Din deutches institut für normung e. v., prüfung von drucken and druckfarben, teil 2: An-

forderungen an die messanordnung von farbdichtemessgeräten und ihre prüfung, ref. nr.

din-16536-2, 1995.

Practice for evaluating the relative lightfastness and weatherability of printed matter, astm

d3424-11 standard, 11 2011. www.astm.org.

U. Agar and J. P. Allebach. An iterative cellular ynsn method for color printer characterization.

In Proc. 6th IS&T Color imaging conf., 1998. pp. 197-200.

J. D. Auslander and R. A. Cordery. Fluorescent hidden indicium. 10 2003. US Pat 7,422,158.

J. D. Auslander, R. A. Cordery, and C. Zeller. Method and system for validating a security

marking. 04 2002. US Pat 7,536,553.

R. Bala. Optimization of the spectral neugebauer model for printer characterization. In J.

Electronic Imaging, Vol. 8, No. 2, 1999. pp. 156-166.

R. Bala. Digital Color Imaging Handbook. Ed. G. Sharma, Chapter 5, Device Characterization,

Section 5.4.5, Lattice-based interpolation, 2003a. pp. 301-304.

R. Bala. Digital Color Imaging Handbook. Ed. G. Sharma, Chapter 5, Device Characterization,

Section 5.2.3, Input device calibration and characterization, 2003b. pp. 275-281.

R. Bala and E. Dalal. A method for quantifying the color gamut of an output device. In

Proceedings of SPIE, Vol. 3018, 1997. pp. 110-116.

R. Bala, R. Eschbach, and Y. Zang. Substrate fluorescence: Bane or boon? In Proc. 15th IS&T/SID

Color Imaging Conference, Albuquerque, NM, 2007. pp. 12-17.

F. Bernardini, J. Mittleman, H. Rushmeier, C. Silva, and G. Taubin. The ball-pivoting algorithm

for surface reconstruction. In IEEE Trans. Visualization and Computer Graphics, Vol. 5, No.

4, 1999. pp. 349-359.

D. H. Brainard, D. G. Pelli, and T. Robson. Display Characterization. 2rd ed., 2002. pp. 172-188.

L. Brehm and H. Erbar. Halftone image produced by printing. 07 2001. US Pat. Appl.

10/482,892.

77

Bibliography

T. Bugnon, Brichon M., and R. D. Hersch. Simplified ink spreading equations for cmyk halftone

prints. In Proc. SPIE, Vol. 6807, 2008. pp. 680717-1 to 12.

Y. Chen, Roy S. Berns, and L. A. Taplin. Six color printer characterization using an optimized

cellular yule-nielsen spectral neugebauer model. In J. Imaging Sci. Technol., Vol. 48, No. 6,

2004. pp. 519-528.

Clapper and J.A.C Yule. The effect of multiple internal reflections on the densities of halftones

prints on paper. In Journal of the Optical Society of America, Vol. 43, 1953. pp. 600-603.

S.A. Connors-Rowe, H.R. Morris, and P.M. Whitmore. Evaluation of appearance and fading of

daylight fluorescent watercolors. In J. Am. Inst. Conserv., 2005. pp. 75-94.

W. J. Coyle and J. C. Smith. Methods and ink compositions for invisibly printed security images

having multiple authentication features. 04 2004. US Pat. Appl. 10/818058.

N. P. Garg, A. K. Singla, and R. D. Hersch. Calibrating the yule-nielsen modified neugebauer

model with ink spreading curves derived from digitized rgb calibration patch images. In J.

Imaging Sci. Technol, Vol. 52, No. 4, paper 40908, 5 pages, 2008.

F. Grum. Colorimetry of fluorescent materials, in Optical Radiation Measurements. F. Grum

and C. J. Bartelson, eds., Academic Press, Vol. 2, Color Measurements, 1980. pp. 235-288.

Karl Guyler. Visualization of expanded printing gamuts using 3-dimensional convex hulls. In

American Ink Marker, Vol. 79, No. 9, September, 2001. pp. 36-56.

C. Haines, S.G. Wang, and K. Knox. Digital Color Imaging Handbook. Ed. G. Sharma, Chapter

6, Digital color halftones, Sections 6.1 to 6.4.11, 2003a. pp. 385-430.

C. Haines, S.G. Wang, and K. Knox. Digital Color Imaging Handbook. Ed. G. Sharma, Chapter

6, Digital color halftones, Sections 6.8.1 to 6.8.9, 2003b. pp. 457-470.

R. D. Hersch and F. Crété. Improving the Yule-Nielsen modified spectral Neugebauer model

by dot surface coverages depending on the ink superposition conditions. R. Eschbach, G.G.

Marcus eds., Vol. 5667, SPIE Color Imaging X: Processing, Hardcopy and Applications, 2005.

pp. 434-445.

R.D. Hersch, Donzé P., and S. Chosson. Color images visible under uv light. In Proc. ACM

SIGGRAPH, ACM Trans. Graphics, Vol. 26, No. 3, 2007. pp. 75:1–75:9.

K. J. Heuberger, Z. M. Jing, and S. Persiev. Color transformations and lookup tables. In Proc.

TAGA/ISCC, 1992. pp. 863-881.

T. Hirokawa, Inui M., T. Morioka, and Y. Azuma. A psychophysical evaluation of a gamut

expansion algorithm based on chroma mapping ii: Expansion within object color data

bases. In NIP23: International Conference on Digital Printing Technologies and Digital

Fabrication, 2007. pp. 175-179.

78

Bibliography

G. Jones II, S. Burke, and P. McDonald. System and methods for product and document

authentication. 07 2002. US Pat. Appl. 10/517299.

Byoung-Ho Kang, J. Morovic, M. Ronnier Luo, and Maeing-Sub Cho. Gamut compression and

extension algorithms based on observer experimental data. In ETRI Journal, Vol. 25, No. 3,

2003. pp. 156-170.

Henry R. Kang. Three-dimensional lookup table with interpolation. In Computational Color

Technology, Chapter 9, Vol. PM159, 2006. pp. 151-159.

Ingmar Lissner and Philipp Urban. Towards a unified color space perception-based. In IEEE

Transactions on Image Processing, Vol. 21, Issue 3, 2012. pp. 1153–1168.

J. Morovic and M. R. Luo. The fundamentals of gamut mapping: A survey. In J. Imaging Sci.

Technol., Vol. 45, No. 3, 2001. pp. 283-290.

Narita and Eto. Method for fluorescent image formation, print produced thereby and thermal

transfer sheet thereof. 2002. US Patent 7,005,166.

T. Ogasahara. Verification of the predicting model and characteristics of dye-based ink jet

printer. In Journal of Imaging Science and Technology, Vol. 48, No. 2, 2004. pp. 130-137.

I. Pobboravsky and M. Pearson. Computation of dot areas required to match a colorimetrically

specified color using the modified neugebauer equations. In Proc. TAGA, Vol. 32, 1972. pp.

65-77.

W. H. Press, B. P. Flannery, S.A. Teukolsky, and W.T. Fetterling. Numerical Recipes. Cambridge

University Press, 1st ed., Section 10.5, 1998. pp. 309-317.

A. Rosenfeld and J. L. Pfaltz. Distance functions on digitla pictures. In Pattern Recognition,

pages 1(1):33–61, 1968.

R. Rossier, T. Bugnon, and R. D. Hersch. Introducing ink spreading within the cellular yule-

nielsen modified neugebauer model. In Proc. IS&T/SID’s 18th Color Imaging Conference,

2010. pp. 295-300.

E. J. Stollnitz, V. Ostromoukhov, and D. H. Salesin. Reproducing color images using custom

inks. In Proceedings of SIGGRAPH 98, ACM, New York, 1998. pp. 267-274.

S.G. Streitel. Fluorescent Pigments (Daylight), Kirk-Othmer Encyclopedia of Chemical Technol-

ogy. 2009. pp. 1-25.

D. Tzeng and R. S. Berns. Spectral-based six color separation minimizing metamerism. In

Proc. 8th IS&T/SID Color imaging conf., 2000. pp. 342-347.

Philipp Urban and Roy S. Berns. Paramer mismatch-based spectral gamut mapping. In IEEE

Transactions on Image Processing, Vol. 20, Issue 6, 2011. pp. 1599-1610.

79

Bibliography

R. L. Van Renessse. Optical Document Security. Artech House, 3rd ed., Chapter 4, Printing Inks

and Printing techniques, 2005. pp. 98-104.

J.A.S Viggiano. Modeling the color of multi-colored halftones. Proc. TAGA, 1990. pp. 44-62.

David R. Wyble and Roy S. Berns. A critical review of spectral models applied to binary color

printing. In Color research and application, Vol. 25, No. 1, 2000. pp. 4-19.

J.A.C Yule and W.J. Nielsen. The penetration of light into paper and its effect on halftone

reproductions. In Proc. TAGA, Vol. 3, 1951. pp. 65-76.

80

A Hue planes of the fluorescent gamutsof both an offset and an inkjet printer

This appendix shows constant hue planes from 0 to 300 degrees between the non-linearly

lightness adapted sRGB gamut and the fluorescent gamut G f of the two considered inkjet

(Figure A.1) and offset (Figure A.2) printers.

81

Appendix A. Hue planes of the fluorescent gamuts of both an offset and an inkjet printer

100

20

40

60

80

L*

hue angle of 0°

20 40 60 800

C *100

100

20

40

60

80

L*

hue angle of 60°

20 40 60 800

C *100

G'sRGB

GfGf

G'sRGB

100

20

40

60

80

L*

hue angle of 120°

20 40 60 800

C *100

Gf

G'sRGB

100

20

40

60

80

L*

hue angle of 180°

20 40 60 800

C *100

Gf

G'sRGB

100

20

40

60

80

L*

hue angle of 240°

20 40 60 800

C *100

G'sRGB

Gf

100

20

40

60

80

L*

hue angle of 300°

20 40 60 800

C *100

G'sRGB

Gf

inkjet

Figure A.1: Comparison of constant hue planes between the fluorescent gamut of an inkjetprinter G f (solid lines) and the non-linearly lightness adapted sRGB gamut G ′

sRGB (dashedlines).

82

20 40 60 800

C *100

100

20

40

60

80

13

L*

hue angle of 0°100

20

40

60

80

13

L*

hue angle of 60°

20 40 60 800

C *100

G'sRGB G'sRGB

Gf

Gf

20 40 60 800

C *100

100

20

40

60

80

13

L*

hue angle of 120°100

20

40

60

80

13

L*

hue angle of 180°

20 40 60 800

C *100

G'sRGB

G'sRGB

Gf Gf

100

20

40

60

80

13

L*

hue angle of 240°

20 40 60 800

C *100

100

20

40

60

80

13

L*

hue angle of 300°

20 40 60 800

C *100

G'sRGBG'sRGB

GfGf

offset

Figure A.2: Comparison of constant hue planes between the fluorescent gamut of an offsetprinter G f (solid lines) and the non-linearly lightness adapted sRGB gamut G ′

sRGB (dashedlines).

83

B Curriculum Vitæ

Romain Rossier is PhD student at the Peripheral Systems laboratory (Ecole Polytechnique

Fédérale de Lausanne, or EPFL) in Lausanne, Switzerland. His research interests include color

prediction, mathematical modeling of printing processes, color printing and optical document

security. He obtained is master degree in computer science from EPFL in 2007. He received

the MERL Best Student Paper Award at the IS&T/SID 19th Color and Imaging Conference

(CIC), San Jose, CA, Nov. 2011 for his paper "Hiding Patterns with Daylight Fluorescent Inks".

He his author and co-inventor in US patent applications.

Personal bibliography

R. Rossier, T. Bugnon and R.D. Hersch. Introducing ink spreading within the cellular Yule-

Nielsen modified Neugebauer model. In Proc. 18th IS&T/SID Color Imaging Conference, 2010.

pp. 295-300.

R. Rossier and R.D. Hersch. Hiding patterns with daylight fluorescent inks. In Proc. 19th

IS&T/SID Color Imaging Conference, 2011. pp. 223-228.

R. Rossier and R.D. Hersch. Gamut expanded halftone prints. In Proc. 20th IS&T/SID Color

Imaging Conference, 2012. to be published.

R. Rossier and R.D. Hersch. Reproducing color images by combining classical and daylight

fluorescent inks. submitted to the IEEE Transactions on Image Processing.

R. Rossier and R.D. Hersch. Synthesis of authenticable luminescent color halftone images. 08

2010. US Pat Appl. 12/805872.

R. Rossier and R.D. Hersch. Ink-dependent n-factors for the Yule-Nielsen modified spectral

Neugebauer model. In Proc. IS&T Fifth European Conference on Color in Graphics, Imaging

and Vision (CGIV), 2010. pp. 202-210.

R. Rossier and R.D. Hersch. Calibrating the ink spreading curves enhanced Yule-Nielsen

modified spectral Neugebauer model with the two-by-two dot centering printer model. In

85

Appendix B. Curriculum Vitæ

Proc. SPIE Color Imaging XIII: Processing, Hardcopy, and Applications, Vol. 7241, 2009. pp.

72411B1-72411B10.

V. Babaei, R. Rossier and R.D. Hersch. Reducing the number of calibration patterns for the

two-by-two dot centering model. In Proc. SPIE Color Imaging XVII: Displaying, Processing,

Hardcopy, and Applications, Vol. 8292, paper 829208, 2012. pp. 1-9.

86

framework for printing with daylight fluorescent inks · 2018. 7. 17. · red, green and blue srgb...

Documents