Master Erasmus Mundus in

Color in Informatics and Media Technology (CIMET)

Spectral Print Reproduction - Modeling and Feasibility

Master Thesis Report

Presented by

Kristina Marijanovic

and defended at

Gjøvik University College

Academic Supervisor(s): prof.Jon Yngve Hardeberg

Jury Committee: Eva Valero - University of Granada, Spain

Joni Orava - University of Eastern Finland
Spectral Print Reproduction - Modeling and Feasibility

Kristina Marijanovic

2012/07/15
 Spectral Print Reproduction - Modeling and Feasibility

Abstract

Towards the goal of spectral print reproduction a few obstacles can be found comparing to repro-
duction of colorimetric match. Human visual system is able to sense wide spectrum of colors and
placing that range on substrate, weather paper or any other, requires skills. Traditional printing
reproduction is built on metamerism phenomena and now the new challenge to overcome that
constraint is set. To create the communication with print mechanism is very important and for
that reason spectral printer modeling represents the crucial step. With this process characteriza-
tion of the printer is done. This depends on the data given and optimal way of its use needs to
be established. With forward printer model the relation between the colorant combination and
the resulting spectral reflectances printed on substrate is defined. The inverse process enables
printer control to define the colorant separation from spectral refelectances to precise values.
Very important step in the whole process is not only measuring the printed material but over-
coming time and resources waste with estimating spectral reflectances and combining them with
already measured ones. Several models were tested by now and our contribution is consisted
of employing Kubelka-Munk theory for estimation step before testing it with cellular model that
makes printed cell smaller than unity and therefore more precise. Among the rest possible theo-
ries for estimation Kubelka-Munk theory was chosen due to its consideration of light scattering.
This was mainly because the study samples used for spectral reproduction were garnament prod-
ucts. Before the target spectral reflectances values can be printed the feasibility of the process
should be considered. Possibility for making that process as much automated as it can be was the
main motivation in creation of feasibility of spectral reproduction algorithm.

i
 Spectral Print Reproduction - Modeling and Feasibility

Preface

I would like to thank you my supervisor, professor Jon Yngve Hardeberg, for not only academic
support but for constant inspiration and all the question marks created above my head after
Monday meetings. Thank you. Next I would like to thank all the members of The Norwegian
Colour and Visual Computing Laboratory, especially to Aditya Suneel Sole and Peter Nussbaum
for their help in the lab as well as friendly talks and interest in my topic. My work on thesis
would never be the same without my colleagues and CIMET friends with whom I have spent four
amazing semesters. I thank you all guys for being always (t)here for me.

In the end I want to express my huge gratitude to my parents, Mirjana and Jure, for always
respecting my freedom of choice and personal space. For this courage I will always admire you
and being thankful to have you in my life.

I also want to thank to Norwegian weather for assuring me there is no such thing as impos-
sible.

ii
 Spectral Print Reproduction - Modeling and Feasibility

Contents

Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Contents . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . v
1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1 Problem statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Answer to the problem stated . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
2 Spectral printer models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.1 Print reproduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Halftoning process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.3 Murray-Davies model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.4 Dot gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4.1 Mechanical dot gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4.2 Optical dot gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Yule-Nielsen model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.6 Dot areas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.6.1 Theoretical area coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.6.2 Effective area coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.7 Inverse Murray-Davies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.8 Spectral Neugebauer model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.9 Demichel model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.10 Yule-Nielsen Modified Spectral Neugebauer Model . . . . . . . . . . . . . . . . . 10
2.11 Cellular spectral Neugebauer Model . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.12 Cellular Yule-Nielsen spectral Neugebauer model . . . . . . . . . . . . . . . . . . 12
2.13 Kubelka-Munk . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Experimental setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1 Print technologies . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.1 Laser print technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.1.2 Crystal point print technology . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.1.3 Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Instrumentation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.3 Metrics of measured and tested patches . . . . . . . . . . . . . . . . . . . . . . . 17
3.4 Estimation of Neugebauer Primaries . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.5 Spectral modeling of the Oce ColorWave600 printer . . . . . . . . . . . . . . . . . 19
3.6 Xerox printer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
4 Spectral printer modeling - results . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

iii
 Spectral Print Reproduction - Modeling and Feasibility

4.1 Kubelka-Munk theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24

4.2 Spectral Neugebauer model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 Yule-Nielsen Modified Spectral Neugebauer Model . . . . . . . . . . . . . . . . . 26
4.4 Cellular spectral Neugebauer Model . . . . . . . . . . . . . . . . . . . . . . . . . 27
4.5 Cellular Yule-Nielsen spectral Neugebauer model . . . . . . . . . . . . . . . . . . 28
5 Feasibility of spectral print reproduction . . . . . . . . . . . . . . . . . . . . . . . . 30
5.1 Collaboration with retailer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
5.2 Color Gamut . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
5.3 Problem specification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
5.4 ICC3D . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
5.5 Proposed Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
5.6 HP Heavy Coated paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
5.7 HP Professional Matte Canvas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.8 HP Artist Matte Canvas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
5.9 HP Universal Coated Paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.10 HP Recycled Bond Paper . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
5.11 Comparing results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
7 APPENDIX A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
8 APPENDIX B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
9 APPENDIX C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

iv
 Spectral Print Reproduction - Modeling and Feasibility

List of Figures

1 Color mixing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Moire effect . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
3 Screening . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
4 Rosettes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
5 Principle of the dot gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
6 Mechanicla dot gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
7 Optical dot gain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
8 Cellular model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
9 Linear testchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
10 CMYK testchart . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
11 Oce - cyan measured and estimated ramp . . . . . . . . . . . . . . . . . . . . . . 19
12 Oce - magenta measured and estimated ramp . . . . . . . . . . . . . . . . . . . . 20
13 Oce - yellow measured and estimated ramp . . . . . . . . . . . . . . . . . . . . . 20
14 Oce - black measured and estimated ramp . . . . . . . . . . . . . . . . . . . . . . 20
15 Oce effective colorant coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
16 Xerox - cyan measured and estimated ramp . . . . . . . . . . . . . . . . . . . . . 22
17 Xerox - magenta measured and estimated ramp . . . . . . . . . . . . . . . . . . . 22
18 Xerox - yellow measured and estimated ramp . . . . . . . . . . . . . . . . . . . . 22
19 Xerox - black measured and estimated ramp . . . . . . . . . . . . . . . . . . . . . 23
20 Oce effective colorant coverage . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
21 Oce measured and estimated NPs . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
22 Xerox measured and estimated NPs . . . . . . . . . . . . . . . . . . . . . . . . . . 25
23 SN model for Oce ColorWave 600 printer . . . . . . . . . . . . . . . . . . . . . . . 25
24 SN model for Xerox Phaser 6250 printer . . . . . . . . . . . . . . . . . . . . . . . 26
25 Yule-Nielsen Modified Spectral Neugebauer model for Oce ColorWave 600 printer 26
26 Yule-Nielsen Modified Spectral Neugebauer model for Xerox Phaser 6250 printer 26
27 Cellular spectral Neugebauer Model for Oce ColorWave 600 printer . . . . . . . . 27
28 Cellular spectral Neugebauer Model for Xerox Phaser 6250 printer . . . . . . . . 27
29 Cellular Yule-Nielsen spectral Neugebauer model for Oce ColorWave 600 printer . 28
30 Cellular Yule-Nielsen spectral Neugebauer model for Xerox Phaser 6250 printer . 28
31 Datacolor SF600 spectrophotometer . . . . . . . . . . . . . . . . . . . . . . . . . 33
32 Lab values of retailer’s color samples . . . . . . . . . . . . . . . . . . . . . . . . . 33
33 Xerox printer gamut . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
34 Xerox printer gamut . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
35 Scheme of the algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
36 Paper A mapping results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36

v
 Spectral Print Reproduction - Modeling and Feasibility

37 Histogram of resulting CMC difference for paper A . . . . . . . . . . . . . . . . . 37

38 3D visualization of paper A gamut . . . . . . . . . . . . . . . . . . . . . . . . . . 37
39 Gamut mapping to A10UL3000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
40 Paper B mapping results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
41 Histogram of resulting CMC difference for paper B . . . . . . . . . . . . . . . . . 39
42 3D visualisation of paper B gamut . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
43 Gamut mapping to B10UL3000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
44 Paper C mapping results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
45 Histogram of resulting CMC difference for paper C . . . . . . . . . . . . . . . . . 40
46 3D visualization of paper C gamut . . . . . . . . . . . . . . . . . . . . . . . . . . 40
47 Gamut mapping to C10UL3000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
48 Paper D mapping results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
49 Histogram of resulting CMC difference for paper D . . . . . . . . . . . . . . . . . 42
50 3D visualisation of paper D gamut . . . . . . . . . . . . . . . . . . . . . . . . . . 42
51 Gamut mapping to D10UL3000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
52 Paper E mapping results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
53 Histogram of resulting CMC difference for paper E . . . . . . . . . . . . . . . . . 43
54 3D visualization of paper E gamut . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
55 Gamut mapping to E10UL3000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
56 Histogram of resulting CMC difference for five papers . . . . . . . . . . . . . . . . 44
57 Gamut mapping to E10UL3000 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
58 Gamut mapping to A10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
59 Gamut mapping to A10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
60 Gamut mapping to A10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
61 Gamut mapping to A10A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
62 Gamut mapping to A10F2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
63 Gamut mapping to B10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
64 Gamut mapping to B10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
65 Gamut mapping to B10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
66 Gamut mapping to B10A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
67 Gamut mapping to B10F2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
68 Gamut mapping to C10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
69 Gamut mapping to C10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
70 Gamut mapping to C10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
71 Gamut mapping to C10A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
72 Gamut mapping to C10F2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
73 Gamut mapping to D10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
74 Gamut mapping to D10D65 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
75 Gamut mapping to D10A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
76 Gamut mapping to D10C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
77 Gamut mapping to D10F2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
78 Gamut mapping to E10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

vi
 Spectral Print Reproduction - Modeling and Feasibility

79 Gamut mapping to E10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

80 Gamut mapping to E10D50 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
81 Gamut mapping to E10A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
82 Gamut mapping to E10F2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
83 CIELAB values of the target and gamut of paper A . . . . . . . . . . . . . . . . . . 51
84 CIELAB values of the target and gamut of paper B . . . . . . . . . . . . . . . . . . 52
85 CIELAB values of the target and gamut of paper C . . . . . . . . . . . . . . . . . . 53
86 CIELAB values of the target and gamut of paper D . . . . . . . . . . . . . . . . . . 54
87 CIELAB values of the target and gamut of paper E . . . . . . . . . . . . . . . . . . 55

vii
 Spectral Print Reproduction - Modeling and Feasibility

1 Introduction

Spectral print reproduction is challange of modern print world and within this master thesis
work two main questions were aimed to be answered. Before any of those are stated it should
be noted that in our work the target object to be spectraly print was ganament product. This is
important because that fact requires scattering of light to be strongly considered.

1.1 Problem statement

Due to statment above the first part of thesis is a try to create the optimal characterisation pro-
cess and model the printer spectrally. Not only measured spectral reflectances of the printed
product was considered but the estimated and mixed measured and estimated reflectances as
well. Prediction of the printed ramps is important to be as much precise due to computational
manipulation of the data. It is not always possible to actually print out and measure all the com-
binations printed. This is due to the exponential problem that raises with number of colorants
being increased. Taking into account importance of the scattering of light in garnament products
the Kubelka-Munk theory was chosen to be employed for estimation of the spectral reflectances.
The second part of the thesis work was dedicated to answering feasibility question. From the
world that surrounds us human visual system is able to sense much wider range of color than
any output system is able to reproduce. Printing machines are part of that system with their own
defined environment and communication channels and the real challenge is these to be over-
come. Certain data with different kind of sensors describes the objects, in our case that data is
defined with the spectrophotometers. Before sending any data to printing process in order to
obtain spectral print the feasibility of that process should be confirmed. If neglecting that fact,
hit-and-miss tries would yield big waste of time and energy. For that reason the automated pro-
cess in verifying the feasibility of color sensation being printed spectrally of certain object was
aimed to be created.

1.2 Answer to the problem stated

In our work big part of experimental process was conducted with the spectral reflectances data
of the garnament samples. This was result of collaboration with the retail store from the USA.
The final goal of retailer is to have the color values obtained under certain condition (geometry,
illuminants) produced on paper substrate in printed form. For the scattering importance of the
material, the Kubelka-Munk was employed in very similar way as it was done by Assefa in [1]
and Blahova in [2]. Within this thesis this was tried to be extended even more with testing not
only basic spectral printer models but breaking the initial cell (unity) to smaller space. The pro-
cess was started with Murray-Davies spectral print model from where inverse effective colorant
coverage was obtained. Once the effective coverage was available the estimation of the Neuge-

1
 Spectral Print Reproduction - Modeling and Feasibility

bauer Primaries was done. Neugebauer Primaries are set of reflectances made of primaries of
the printer and their possible combinations of full coverage plus the reflectance of the substrate
(paper). Not only the measured and estimated reflectances were used in testing the models but
also their combinations (mixed reflectances). The test was conducted for spectral Neugebauer
model, Yule-Nielsen Spectral Neugebauer Model for two different printers and technologies. The
new step and contribution was introducing the estimation with Kubelka-Munk to cellular version
of mentioned models. That way the cellular spectral Neugebauer model and cellular Yule-Nielsen
Spectral Neugebauer Model were prepared to be tested. Comparing to non-cellular spectral mod-
els cellular extention resulted with more accurate results.
In the second part of thesis where the challenge of spectral print feasibility of target object was
tried to be automated, several different steps in creation of that process were created in one so
that the new algorithm, which includes all the segments in one step that saves time, was pro-
posed. In this way big amount of input data can be initially placed and from the output it can be
seen exact which parts of the gamut for targeted spectral reproduction are missing. In this way
the answer to set of the spectral print feasibility of number of objects, under the same illuminant,
was given.

The thesis work was conducted with laser, inkjet and CrystalPoint (inkjet subset) printing
technologies so the initial title of the master thesis ’Spectral modelling of multi-channel inkjet
printer’ was changed. This way the scope of research was extended to give opportunity as well
to gamut area of spectral print reproduction.

2
 Spectral Print Reproduction - Modeling and Feasibility

2 Spectral printer models

2.1 Print reproduction

Print reproduction is taking advantage of subtractive color mixing. That is the kind of mixing
where colored filters are illuminated with white light from behind, as illustrated in fig. 1. The
commonly used subtractive primary colors are cyan, magenta and yellow. If all three are over-
lapped in effectively equal mixture all the light will be subtracted therefore giving black. Unlike
this additive color mixing creates combination of red, green and blue light on the black back-
ground giving the white if all three overlap. Subtractive type of color mixing is characteristic for
devices that use paints and pigments among as well are printers.

Figure 1: Additive color mixing represented on the left side versus subtractive color mixing on the right side
of the figure.

2.2 Halftoning process

The process of creating screen dots, also known as halftoning [3] [4], converting continuous
tone originals to printable elements that is suitable for printing reproduction. The screen dots
are different sized and that, defending on their position, helps human eye to perceive contin-
uous tone while observing the print. Depending on the printing process, there are usually only
two states of ink transfer (printing or non-printing). This is why the varying visual impression
of lightness or darkness must be created by varying the size of the halftone dot. If the screen
structures are fine enough for the viewing distance of the screened image, the integrating effect
of the human eye will in a way smooth out the image and the observer sees an image that, with
it’s continuous tone gradations, is visually consistent with the original. The greater the number of
screen dots per image area, the more natural the effect of the image . The term screen ruling (or
screen frequency) is used to define how close are the screen dots to each other [5] [6]. When

3
 Spectral Print Reproduction - Modeling and Feasibility

observing a 60 lines/cm screen (i.e., the screen frequency, usually referred to as the screen ruling
L = 60 lines per cm, corresponding to a dot spacing w = 1/L = 0.167 mm) at a normal reading
distance (approx. 30 cm), the eye is usually no longer able to detect the individual dots [?] [7].
With the help of the computer, the basic, manual screening process has been scientifically and
mathematically analyzed in order to develop new, electronic reproduction processes. The basic
principle of breaking down an image into different-sized screen dots which were equally spaced
apart was initially preserved. The first systems capable of electronic screening were output scan-
ners that imaged the films with very finely focused laser beams. The individual, different-sized
dots were assembled from several laser spots (picture elements, pixels) [8] [9]. All laser image
setters now work on this principle, first into the four basic colors (cyan,magenta, yellow, and
black) of the printing process, and then into individual printing elements [10]. The screening of
the individual color separations is done at different screen angles. This is necessary because the
overprinting of different colors will occur without problems only using this procedure. Improper
positioning of color separations causes interference, or so-called Moire patterns. Moire patterns
can severely impair the image impression. This can be easily observed from figure 2.

Figure 2: Moire effect created by interference through overlaying two periodic structures with similar screen
angles - taken from [8].

With conventional reproduction technology, the color separations for the three chromatic colors
were each offset at screen angles of 30 degrees in relation to each other.With four colors the least
distinctive color, yellow, is output at an optimal angle of only 15 degrees in relation to magenta
or cyan that is seen from figure 3.

Even if the optimal angle was formed when attempting to reduce interference effects/Moire phe-
nomena, in homogeneous color areas rosettes may be formed as in figure 4.

4
 Spectral Print Reproduction - Modeling and Feasibility

Figure 3: Angling of the screened color separations with rosette formation - taken from [8].

Figure 4: All color separations are printed with the same color so that Moire effect and rosettes created are
easy to spot - taken from [8].

The formation of a rosette also depends on the relative positioning of the color separations
to one another [8] [11]. Irregular rosette formation during printing can be caused by color
register fluctuations. The finer the screen ruling, the less detectable the distorting structure. In
printing technology most jobs make use of a screen ruling of about 60 lines/cm. Nevertheless,
with originals having a distinctive, fine structure of their own (textile or filigree patterns, for
example) structure-related Moire patterns may occur and are virtually unavoidable. Fine screens
with up to 150 lines/cm are sometimes used for the reproduction of extremely fine structure
details, even though they reduce Moire effects, they do not always prevent them [12]. In case
of fine screens technical process-related requirements on reproduction technology and printing
are exact imposed.

2.3 Murray-Davies model

Spectral printer model is a way to characterize printing process [13] [14]. With different
models we are able to estimate the resulting spectral reflectances on the printing substrate for

5
 Spectral Print Reproduction - Modeling and Feasibility

any colorant combination. In the case of just one colorant print - monocolorant print, Murray-
Davies model is fundamental equation for resulting spectral reflectance estimation. This model is
regression-based model meaning that is relatively simple, with parameters fit to a set of data. It
is useful for modeling printer output because it tends to be reasonably accurate and its simplicity
allows for short calculation times [15]. Furthermore, having regression-based characteristics this
model emulates the behavior of the system. By the assumptions of additivity inside a halftone
cell, the spectral reflectance factor of a halftone cell printed on paper by a single colornat can be
estimated based on the Murray-Davies equation [16]. In order to predict the reflectance of a sin-
gle colorant coverage Murray-Davies model is employed for estimation of the spectral reflectance
of a colorant coverage c as follows:

^r(λ) = (1 − c) × rpaper (λ) + c × rcol,max (λ) (2.1)

where rpaper is the paper measured spectral reflectance, rcol,max is the measured spectral re-
flectance of the paper covered by the colorant at maximum coverage. Usually rcol,max value
is 100% and ^r(λ) is the predicted spectral reflectance. Simple perform of linear interpolation
between the reflectance of the paper and that of the full coverage colorant is result of Murray-
Davies model. The estimated reflectance is the sum of the weighted reflectance of substrate
and colorant where the weights are the distances along the line in colorant space. Estimated
spectral reflectances obtained by Murray-Davies model assumes linear relation for each colorant
printed on substrate. Linearity here denotes that for each primary the reflectance factors should
be summed up according to their coverage dot areas. That is, if the coveragre (fraction) dot ar-
eas of a primary are summed up to a percent, from 0% to 100%, then the resultant reflectance
factor is summed up to a percent of the reflectance factor of the primary at 100% area coverage.
In physical print this relation is ruined and fails due to phenomena that limits linearity, this is
known as dot-gain.

2.4 Dot gain

Dot gain is phenomenon that causes printed material to look darker than initially wanted [17]
[18]. This is due to diameter of halftone dots which during printing process increases. Increase
of the halftone dot placed on the substrate is caused by the optical and physical properties of the
media and printer employed both in prepress and printing process. There are two sources of dot
gain, mechanical and optical dot gain.
2.4.1 Mechanical dot gain
Mechanical dot gain is caused by printing ink submerging and spreading when delivered onto
a paper substrate causing the physical dot size to be changed. This is due to the physical ink
spread. A more fluid ink would spread more resulting in higher dot gain. Absorbance of the
substrate also affects dot gain [19] [20]. Higher absorbance of the substrates is the higher is
dot gain cause by easier spreading of the colorant through the paper substrate.
2.4.2 Optical dot gain
Optical dot gain comes from light flux that penetrates the colorant film, enters the paper sub-
strate, scatters and finally emerges from non-ink area. In the same way light flux can enter the

6
 Spectral Print Reproduction - Modeling and Feasibility

Figure 5: During printing process several phenomena occur that makes resulting printed dot bigger in size
than initially wanted. Result of it is called dot-gain.

Figure 6: Mechanical dot gain occurs when printing ink is submerging and spreading when delivered on a
paper substrate causing the physical dot size to be changed – taken from [21].

paper substrate and get scattered, penetrate the colorant film and finally emerge from there.
This is basically resulted by the interaction of light with the colorant and paper. As the light
penetrates the colorant on the way to the paper substrate, a certain amount of it is diffused. This
diffused light may leave the edge of the placed colorant dot in different ways: to reflect light
beyond physical dimensions of the placed dot or get trapped underneath the dot. In this way,
even they are physical different, the dot appears bigger in size. These two physical phenomena
are the reason for printing models having problem in prediction of dot size which has significant
influence to estimated spectral reflectances.

2.5 Yule-Nielsen model

Towards better and more accurate printer modeling, dealing with both mechanical and optical
dot-gain, n-factor was employed to modify initial Murray-Davies relation. This was introduced
by Yule and Nielsen who worked on light penetration and scattering [22]. Their theoretical
analysis showed that the nonlinear relationship between measured and predicted reflectance can
be described with a power function. Starting from the base Murray-Davies model in equation 2.1

7
 Spectral Print Reproduction - Modeling and Feasibility

Figure 7: Optical dot occurs in the process of interaction of light with the ink and substrate.

they added the exponent 1/n to the reflectance values:

1 1 1
^r n (λ) = (1 − c) × rpaper
n
(λ) + c × rcol,max
n
(λ) (2.2)

Even though their n − factor is an empirical parameter through different research it was shown
that it is capable to maintain high accuracy of model fitting. For this reason Yule-Nielsen n −
factor is still widely employed for its simplicity. In order to find the optimal n factor iterative
process which estimates the spectral reflectances for various n factor values can be employed.
Next, by comparing the estimated spectral reflectances with the measured spectral reflectances
more information will be obtained so that the n − value which gives the minimal colorimetric
differences or smallest spectral difference can be chosen.

2.6 Dot areas

During the printing process paper substrate in close interaction with colorants results in different
printed area than initially set up with digital values. This is mainly due to mechanical dot gain
previously described in 2.4.1. Awareness of initial values being changed in final printed material
is very important and distinction of those various types should be made. There are two different
areas that are covered by colorant dots: theoretical area and effective area.
2.6.1 Theoretical area coverage
Theoretical area coverage is calculated from the actual binary image sent to the printer [23]
[24]. While not always physically reasonable, a square pixel shape is assumed, and theoretical
area coverage is determined as shown by the percentages. The simple way of describing the-
oretical coverage are pure numerical values or steps in ramp of a colorant. For instance, span
from 0% to 100% coverage in step of 5% will result in ramp of 20 sub-patches. These values are
always predeterminated and can be easy visualized i.e. with various test charts.

8
 Spectral Print Reproduction - Modeling and Feasibility

2.6.2 Effective area coverage

Opposite to that effective dot area (effective area coverage) is always an estimated value. For
example, if the spectral reflectance is accurately predicted by a particular area coverage, this is
referred to as the effective area coverage, aeff . This is equivalent to selecting the best propor-
tion such that it most closely matches the measured spectral reflectance at area coverage. It is
important to establish relation between theoretical and effective area coverage [25]. This value
can be obtained by calculating aeff for each spectral reflectance measurement corresponding to
a known theoretical colorant value sent to the printer. After being established, this relation is
usually expressed in the form of a look-up table (LUT) for single colorant printing but can be,
and in most of cases it is, performed for all the available colorants of a printing system.

2.7 Inverse Murray-Davies

Estimation of effective dot area can be done by modified use of the Murray Davies model. By
rearranging 2.1 predicted effective dot area covered by the colorant of a known input value,
which is in this case theoretical colorant coverage, can be found as follows:
rλ=min,meas − rλ=min,s
aeff = (2.3)
rλ=min,t − rλ=min,s
It can be noticed that here ^r is replaced by the measured reflectance. Also, reflectance values
are subscripted λ = min that means this calculation is being performed at a single wavelength.
aeff can be seen as the factor by which the solid ink reflectance should be scaled to equal the
measured reflectance of the sample. This equation is very important in fundamental stage of
modeling because it can be used to estimate dot gain, therefore improving the estimation of area
coverage. Since our data requires not only calculation with single reflectance but whole spectra
range it should be extended to a spectral form. In order to that aeff values needs be determined
by a matrix calculation using least squares analysis:

aeff = Rmeas,adj RTt,adj (Rt,adj RTt,adj )−1 (2.4)

Reflectance terms in this equation are row vectors, the length of which is the number of wave-
lengths in the spectral measurement. Here, Rmeas,adj = Rmeas − Rs and Rt,adj = Rt − Rs , and
the superscipts T and -1 indicate matrix transpose and inverse [26] [27] . This calculation needs
to be repeated for each patch in the separation ramp.

2.8 Spectral Neugebauer model

Apart for models being accurate as possible considering mechanical and optical dot gain tendency
to developing one for not only monochrome but multicoloroant print was introduced by Hans E.J.
Neugebauer in 1937. Neugebauer model considers all the channels of colorant of a printer and
by extending initial Murray-Davies model gives a precise spectral reflectance value of possible
combinations of colorants by taking into account 100% (full) colorant coverage [28]. By this
model it is expressed that in case of k colorant printer (k number of channels) there will be
2k Neugebauer Primaries(NPs). Set of printer’s NP is created with spectral reflectances of full
coverage of primaries (k), spectral reflectances of their full coverage overlaps and reflactances of

9
 Spectral Print Reproduction - Modeling and Feasibility

the paper substrate. Neugebauer computes the predicted reflectance vector ^r as the sum of each
combination of colorants (NPs) weighted by their coverage and the spectral Neugebauer (SN)
model can be then written as follows:
2X
k
−1
^r(λ) = ai Ri (λ) (2.5)
i=0

where the summation over i represents the Nuegebauer Primaries: paper substrate, single col-
orants separations, and color overlaps; ai is a scalar weighting representing the probability that
a point on the page of the color mixture is covered by the ith NP and Ri (λ) is the reflectance
vector of the primary. Not only for the full colorant coverage estimation of reflectances and re-
sulting combination Neugebauer model was base for estimation the relative area of each primary.
Relative area is the surface of the paper needed the most in everyday printing. It denotes all the
steps on the way to reach full colorant coverage. Since it is not common that prints are made
of just full colorant coverage this was important step. Towards the goal of estimating relative
(fractional) colorant printed area Demichels model was employed.

2.9 Demichel model

The Demichel model is a statistical model that estimates the NP coverage for a given colorant
combination. By this model it is assumed that the colorant coverage is statistically independent as
well as placement between dots is independent. In our work the basic formulation of Demichel
equation was used in extended version to four colorants. For the simplicity, and exponential
problem that occurs with adding colorants, here is shown how Demichel equation works in
basic, two colorant combination:

w0 = (1 − c1 )(1 − c2 )
w1 = c1 (1 − c2 )
(2.6)
w2 = (1 − c1 )c2
w12 = c1 c2

where w0 is the weight for the paper, since no colorant is placed on paper substrate. w1 is the
weight of the first colorant c1 , w2 for colorant c2 and w12 for the overlap of colorant c1 and
colorant c2 . It can be noted that in case of two colorants there will be 22 weights. For making the
further calculation easier the obtained weights are represented in a vector form W as follows:

W = [w0 w1 w2 w12 ] (2.7)

2.10 Yule-Nielsen Modified Spectral Neugebauer Model

Not only internal reflections in paper but other phenomena affects reflection spectrum of colorant
being printed at paper substrate. As mentioned before the nonlinear relationship between the
reflection spectra of paper and colorant placed was obtained empirically by Yule and Nielsen
[29] [30] [31]. That was determinated by applying a power function, whose exponent n is fitted
according to a limited set of measured patch reflection spectra. This non-linear relationship was

10
 Spectral Print Reproduction - Modeling and Feasibility

Figure 8: Example of cellular model where the reflectance of the test patch X is now calculated from the
reflectance of R01, R02, R11, and R12 - from [34].

applied to the spectral Neugebauer model which resulted with more accurate model the Yule-
Nielsen modified spectral Neugebauer (YNSN) model [32]:
2X
k
−1
1
^r(λ) = (ai Ri (λ) n )n (2.8)
i=0

Yule-Nielsen modified spectral Neugebauer model is still accurate and simple enough to prevent
complex calculation in spectral estimation of print. It is important to note that when fitting
n − value, effective area coverages must be calculated again and then put in the Demichel
equations each time that n − value is changed.

2.11 Cellular spectral Neugebauer Model

With spectral Neugebauer model the whole gamut of the printer has been achieved by interpola-
tion from a few points on its surface [33]. This creates certain space of data. Taking into account
spreading of the light as well as non-linear relationship it can be easily noted that interpolation
of that data space would yield with result far away from accurate result. Besides applying the
YuleNielsen n-value correction to deal with the nonlinearity and given actual area coverage ef-
fective area coverage there was another try in to creating that data space smaller. That approach
is aiming to reduce the space over which interpolation is performed by providing more primaries
that are inside the printer gamut. This can be done by measuring each colorant at 0, 50, and
100% area coverage so that the printer gamut is subdivided into smaller cells. This approach,
because of dividing the initial gamut space into smaller cells, was called Cellular Neugebauer
and was first presented by Heuberger et al.
Here a simplified case is presented graphically, with two divisions per separation, results in four
cells for two-color printing. When using non-cellular Neugebauer, reflectance of the test patch,

11
 Spectral Print Reproduction - Modeling and Feasibility

marked with an X, is predicted by interpolating between R00, R02, R20, and R22 8. The use of
cellular Neugebauer decreases the size of the interpolation cell [34]. The reflectance of the test
patch is now calculated from the reflectance of R01, R02, R11, and R12 8. If more accuracy is
required, more nods (primaries) can be added at the expense of computation and measurement
time. When using cellular Neugebauer, the interpolation is identical to 2.5. However, additional
computation is required to determine which primaries to use, or equivalently, selecting in which
cell the interpolation is to be performed [35] [36]. This is an iterative process. Simiral to this,
again search for rest colorant primaries is needed. With the identified cell, the effective area
coverage of the colorant is calculated. Equations employed here are analog to 2.3 as well as the
least square analysis form 2.4. They show the scalar form and the least squares analysis on the
spectral reflectance data.
The main difference in terms relating to the cellular case involves following: adjusted values are
those that have had the reflectance of the lower primary subtracted. So, here aeff represents
the effective colorant area coverage that is used to calculate the effective area coverages of the
relevant Neugebauer primaries [37] [38]. This can be applied to predict spectral reflectance.
What is also important to observe here is that cell boundaries do not need to be spaced uniformly.
The best is to sample the space more frequently in areas where reflectance changes rapidly with
respect to input area coverage but this depends on the particular printer used.

2.12 Cellular Yule-Nielsen spectral Neugebauer model

Another useful extension of the cellular case is the addition of a Yule-Nielsen factor [39]. The
accuracy of the cellular Yule-Nielsen spectral Neugebauer model (CYNSN) improves in general
the accuracy of the plain YNSN model since the physical printer transfer function is sampled on
more than the 2k Neugebauer primaries for a k colorant printer. The main drawback of dealing
with this model is the need of more measurements. For that reason it is necessary to find a good
compromise between measurement effort and spectral accuracy.

2.13 Kubelka-Munk
The Kubelka-Munk equation defines a relationship between spectral reflectances of the sample
and its absorption and scattering characteristics of the samples with opacities greater than 75%
using the equation:
K (1 − 0.01R(λ))2
= (2.9)
S 2(0.01R(λ))
Where R is the reflectance given in percent and K and S are the absorption and scattering coef-
ficients of the colorant. These ratios are calculated for all colorants separately and then used for
computing the K/S ratio of the mixture where those colorants are used. This can be done just by
extending the above Kubelka-Munk theory. This is represented as following:
)
K(λ) K(λ) K(λ) ( 1
^r(λ) = 1 + − [( 2+2 ] ) (2.10)
S(λ) S(λ) S(λ) 2

12
 Spectral Print Reproduction - Modeling and Feasibility

Extending the equation further K/S of the mixture is given by the sum of the K/S values of the
individual colorants as it follows:
       
K(λ) K(λ) K(λ) K(λ)
=a +b + ... + (2.11)
S(λ) mixture S(λ) colorant1 S(λ) colorant2 S(λ) paper

where a, b, c are the concentrations of the colorants in the mixture.

Finally, value of K/S ratio for all colorants and their combinations is then used in obtaining the
reflectances of the primaries and their combinations as follows:
s
 2
K(λ) K(λ) K(λ)
R∞ (λ) = 1 + − +2 (2.12)
S(λ) S(λ) S(λ)

From the given equations by Kubelka-Munk it can be noticed that this method contains char-
acteristics required in our calculations. Not that is only presented in a very simple mode but it
also deals with the scattering problem in the substrate [40]. This consideration is important in
our calculations of estimating the Neugebauer Primaries and it will be shown within following
chapters.

13
 Spectral Print Reproduction - Modeling and Feasibility

3 Experimental setup

3.1 Print technologies

In our experimental part two four-channel printers were used for spectral modeling. We had
twelve colorant printer channel on dispose(HP Designjet Z3200) but the four-channel ones were
still chosen over it. The reason to that was control of the twelve-colorant printer that could not
be done with exact precision. For both four channel printers we were able to control all of the
channels, cyan, magenta, yellow and black. These two printers are in the same time represents
of two different print technologies. Xerox Phaser 6250 use laser print technology while Oce
ColorWave600 uses CrystalPoint technology.
3.1.1 Laser print technology
Laser printers rely on the same technology that is used in photocopying machines. This process
is known as electrophotography. Electrophotographic process in laser printers, involves six basic
steps:
1. A photosensitive surface (photoconductor) is uniformly charged with static electricity by a
corona discharge.
2. The charged photoconductor is exposed to an optical image through light to discharge it
selectively and forms a latent or invisible image.
3. Development is done by spreading toner, a fine powder, over the surface, which adheres only
to the charged areas, thereby making the latent image visible.
4. An electrostatic field transfers the developed image from the photosensitive surface to a sheet
of paper.
5. Transferred image is fixed permanently to the paper, by fusing the toner with pressure and
heat.
6. Cleaning of all excess toner and electrostatic charges from the photoconductor to make it
ready for next cycle.
Laser printers are among the ones that offer the best print quality i.e. the highest resolution. The
unique difference is the method of exposition or formation of the latent image. These printers
rely on a laser beam and scanner assembly to form a latent image on the photo-conductor bit
by bit. The scanning process is similar to electron beam scanning used in CRT. The laser beam
modulated by electrical signals from the printer’s controller is directed through a collimator lens
onto a rotating polygon mirror (scanner), which reflects the laser beam. Then reflected from the
scanner laser beam pass through a scanning lens system, which makes a number of corrections
to it and scans on the photoconductor [8].

14
 Spectral Print Reproduction - Modeling and Feasibility

3.1.2 Crystal point print technology

Oce’s color toner technology utilizes toner pearls instead of liquid ink. Moreover that ink is a solid
TonerPearl, available in CMYK colorants. When the TonerPearl is heated to 300C it becomes a gel.
The gel is jetted onto the plain, untreated paper using an Oce-developed inkjet head and dries
instantly upon hitting the media. Inkjet technology makes use of tiny ink droplets to facilitate
direct printing without the device coming into contact with the printed surface. Because this
technology enables non-contact printing, it can be applied to all kinds of media and is now
being introduced for use in a wide range of fields ranging from general purpose to industrial.
The simple structure that combines the inkjet print head with the scanning mechanism offers
the merit of keeping device costs down. In addition, since they require no plates, inkjet printers
offer the advantage of major savings in printing set up time compared to conventional printing
systems that require a fixed printing block or plate, etc. (e.g. screen printing). Since it dries right
away, there is no wicking and no spreading of ink. This way the print obtained is sharped in
lines, high readability of fine details, and smooth, even area fills. Further, the solid crystals add
a silk shine on plain paper.
3.1.3 Experiment
For our experiment we have used two CMYK printers for spectral modeling, Xerox Phaser 6250
color laser printer and HP 1220C DeskJet printer. Both printers are a 4 channel (CMYK) printers.
The colorant combinations have been printed on different types of paper. For laser print staple
copy paper (80g/m2 ) was used, this was sheet fed printer. In other printer, using crystal point
technology for printing, Red Label paper, of gramature 80g/m2 , was printed on. This was roll
though this printer is as well sheet fed printer. For obtaining data used in our spectral printer
modeling calculation test chart Linear CMYK i1 i0 P.M. 5.0.7 was used. It is created of all together
128 patches, containing four different color ramps, one per primary channel. Also, one ramp
consist of 30 sub-patches or thirty steps gradually increasing to final patch that is full coverage
of primary colorant. This test chart was suitable for our work because it has all four channels
ramps of the primaries in printers being spectrally modeled. This is represented with figure 9 For

Figure 9: Testchart that is created with linear steps in amount of colorant being placed on paper substrate.
This way ramps of printer’s primaries were created.

visualizing and printing the chart we used Adobe Photoshop CS5. The main reason to that option
was to assure that no color management will be employed while printing process. This was done
by testchart opened in CMYK mode and since the testchart is actual TIFF file, we were able to
find pure and independent primary prints by setting both monitor and printer color management
off. In example, magenta patch of the testchart contains only magenta printer colorant after a

15
 Spectral Print Reproduction - Modeling and Feasibility

Figure 10: Testchart that is created with four channels and their combinations. This testchart was used for
Neugebauer primaries inputs and later as test material in modeling.

print being made. This was the case for both the printers. This way it was possible to be sure that
the rest colorants will be printed independent like magenta was. It was important to us to verify
this step to make sure that the primaries of our printers are independent of each other and there
will not be any unnecessary mixing of colorants in printing ramps of pure channels.

3.2 Instrumentation
Among the spectrophotometers in our laboratory GretagMacbeth Spectrolino was chosen to be
used in measuring spectral reflectance of the printed testcharts. Advantage of this instrument
is that it can be positioned exactly on color and paper charts for accurate measurements not
manually but using software options to control position over the measured sample. To ensure
consistent results, Spectrolino uses an annular lens optical system. This multi-directional ap-
proach to gathering light in a circle enables Spectrolino to read the target at any angle. It also
includes a UV cut filter and has a 45/0 geometry, i.e. with a D50 light source illuminating the
printed sample at an angle 45 degrees and a sensor capturing the reflected spectrum at 0 degrees.
The UV filter was used for removing effects of optical brightness of the papers. To get the exact
spectral reflectance data Spectrolino was combined with measure tool software of Profile Maker
pro 5.0.9 for measuring test charts. For measuring Neugebauer primaries another test chart was
used, IT8.7-3 CMYK i1-i0 PM 5.0.7. presented with figure 10 That chart is created of total 936
patches, primaries and their combination that also includes Neugebauer primaries. Here the au-
tomated work of Spectrolino spectrophotometer was recognized as advantage because it saved

16
 Spectral Print Reproduction - Modeling and Feasibility

us a lot of time comparing to manually placing the instrument over each patch of test chart.

3.3 Metrics of measured and tested patches

Spectral reflectance measurement of the patches was conducted with self backing process. Testcharts
were measured with GretagMacbeth Spectrolino by placing them on the 20 similar blank papers.
Spectral reflectance of printed NPs was obtained that way and then and compared with the
estimated once. Both spectral and colorimetric distance metrics were employed in obtaining fi-
nal results [41]. In that way it can be noticed how much diverse they are both spectrally and
perceptually. For that reason average root mean square error (RMSE) and goodness of fit coeffi-
cient (GFC) between the measurements of spectral reflectances of NPs: measured, estimated and
mixed combination and actual measurements of the 108 colorant combination patches of the
test chart. These 108 patches were extracted from IT8.7-3 CMYK i1-i0 PM 5.0.7 testchart 10.
from field N1 to R12.

The root mean square error (RMSE) for evaluation of spectral estimation was obtain as fol-
lows:
! 21
1X
RMSE = (rm (λ) − re (λ))2 (3.1)
n
λ

where rm (λ) and re (λ) are the measured and estimated spectral reflectances. The n value here
corresponds to the number of the wavelengths. the perfect match of the rm (λ) and re (λ) equals
to zero nad the worst match goes to infinity.
The Goodness-of-Fit Coefficient (GFC) is the multiple correlation coefficient, the square root
of the all re (λ) estimated spectral variance with the respect to the measured original spectral
data of rm (λ) is as follows:

P
rm (λj )re (λj )
j
GFC = ! 21 ! 21 (3.2)
P P
[rm (λj )]2 [re (λj )]2
j j

The GFC is ranged from 0 to 1 and the perfect match equals to 1. GFC value of 0.999 is considered
very good spectral match
.
Average ∆Eab, ∆E94 and ∆E00 for illuminant D50 were also calculated.
∆Eab
q
∗
∆Eab = (L∗2 − L∗1 )2 + (a∗2 − a∗1 )2 + (b∗2 − b∗1 )2 (3.3)
∆E94

17
 Spectral Print Reproduction - Modeling and Feasibility

s 2 2 2
∗ ∗
∆L∗
 
∗ ∆Cab ∆Hab
∆E94 = + + (3.4)
KL 1 + K1 C1∗ 1 + K2 C1∗
where
∆L∗ = L∗1 − L∗2
q
C1∗ = a∗1 2 + b∗1 2
q
C2∗ = a∗2 2 + b∗2 2
∗
∆Cab = C1∗ − C2∗
q q
∗
∆Hab ∗ 2 − ∆L∗ 2 − ∆C∗ 2 =
= ∆Eab ab ∆a∗ 2 + ∆b∗ 2 − ∆Cab
∗ 2

∆a∗ = a∗1 − a∗2

∆b∗ = b∗1 − b∗2
∆E00
s
2 2 2
∆L 0 ∆C 0 ∆H 0 ∆C 0 ∆H 0
  
∆E∗00 = + + + RT (3.5)
SL SC SH SC SH
where
∆L0 = L∗2 − L∗1

L∗1 + L∗2 C∗ + C∗2

L̄ = C̄ = 1
2 2
 s   s 
a C̄7 a C̄ 7
1   a02 = a2 + 2 1 −
a01 = a1 + 1− 
2 C̄7 + 257 2 C̄7 + 257

C01 + C02
q q
C̄0 = and ∆C 0 = C20 − C10 where C01 = C02 =
02 2 02 2
a 1 + b1 a 2 + b2
2

h01 = tan−1 (b1 /a01 ) mod 2π, h02 = tan−1 (b2 /a02 ) mod 2π


 0
h2 − h1
0
|h01 − h02 | ≤ π
∆h 0 = h02 − h01 + 2π |h01 − h02 | > π, h02 ≤ h01

 0
h2 − h01 − 2π |h01 − h02 | > π, h02 > h01

q (h01 + h02 + 2π)/2 |h01 − h02 | > π
∆H0 = 2 C01 C02 sin(∆h0 /2), H̄0 =
(h01 + h02 )/2 |h01 − h02 | ≤ π

T = 1 − 0.17 cos(H̄0 − π/6) + 0.24 cos(2H̄0 ) + 0.32 cos(3H̄0 + π/30) − 0.20 cos(4H̄0 − 21π/60)

18
 Spectral Print Reproduction - Modeling and Feasibility

2
0.015 L̄ − 50 0 0
SL = 1 + q
2 SC = 1 + 0.045C̄ SH = 1 + 0.015C̄ T
20 + L̄ − 50
s "  0 2 !#
C̄ 07 π H̄ − 275◦
RT = −2 sin exp −
C̄ 07 + 257 6 25

3.4 Estimation of Neugebauer Primaries

After measuring exact spectral reflectance data of Neugebauer primaries from the test charts
estimation of the same patches being measured was done. In our experiment estimation of the
NPs reflectances was obtained by using Kubelka-Munk theory mentioned previously explained
in 2.13. Measured reflectances of the primary colorants of the printers were used as a starting
point for the estimation of the reflectance of the rest of the NPs. First, we calculate the K/S
ratios for these primaries. Then we used these ratios in order to compute the K/S ratios for the
combinations of CMYK primaries. In the end, the reflectance of each NPs from their resulted K/S
ratios was computed. The 16 NPs were printed by both printers from IT8.7-3 CMYK i1-i0 PM
5.0.7 testchart, both on different paper substrate.

3.5 Spectral modeling of the Oce ColorWave600 printer

The process of modeling the CrystalPoint technology printer Oce ColorWave600 started with the
measuring the color patches from the Linear testchart 10 where the ramps were given for all the
four channels of printer. The spectral reflectances of measured ramps are given with fig. 11 for
cyan, fig. 12 for magenta, fig. 13 and fig. 14 for yellow and black.

Measured Oce Cyan reflectances − 0 to 100% coverage Oce MD Estimated Cyan reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 11: Oce - measured and estimated spectral reflectances of cyan ramp.

After the Linear testchart was printed the theoretical coverage values were replaced with
the effective coverage values by inverting the Murray-Davies spectral print model. The obtained
effective colorant values confirms the dot gain effect. This can be seen from the fig. 15
Once when the colorant coverage was established from actual printed values the estimation
of the color ramps was started. This needed to be conducted for each of the ramp of the linear

19
 Spectral Print Reproduction - Modeling and Feasibility

Measured Oce Magenta reflectances − 0 to 100% coverage Oce MD Estimated Magenta reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 12: Oce - measured and estimated spectral reflectances of magenta ramp.

Measured Oce Yellow reflectances − 0 to 100% coverage Oce MD Estimated Yellow reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 13: Oce - measured and estimated spectral reflectances of yellow ramp.

Measured Oce Black reflectances − 0 to 100% coverage Oce MD Estimated Black reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 14: Oce - measured and estimated spectral reflectances of black ramp.

20
 Spectral Print Reproduction - Modeling and Feasibility

Oce Cyan Dot Gain Oce Magenta Dot Gain

1 1

Effective coverage

Effective coverage
0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
Theoretical coverage Theoretical coverage
Oce Yellow Dot Gain Oce Black Dot Gain
1 1

Effective coverage
Effective coverage

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
Theoretical coverage Theoretical coverage

Figure 15: Effective colorant coverage of Oce ColorWave 600 printer.

testchart separately since the effective coverage is specific for the each colorant channel. The
fact that the ramps were monochromatic resulted with using Murray-Davies model in process of
estimation. Predicted spectral reflectances are shown in fig. 11 for cyan, fig. 12 for magenta,
fig. 13 and fig. 14 for yellow and black ramp. It can be seen that the estimated reflectances are
more uniform comparing to measured spectral reflectances. In case when the effective coverage
was not considered the uniformity of the spectral reflectances curve is even more expressed. This
is due to dot gain effect and not considering variables as effective coverage or later on, the n
value.

3.6 Xerox printer

The process of modeling the laser technology printer Xerox Phaser 6250 started with the mea-
suring the color patches from the Linear testchart 10 where the ramps were given for all the four
channels of printer. The spectral reflectances of measured ramps are given with fig. 16 for cyan,
fig. 17 for magenta, fig. 18 and fig. 19 for yellow and black.
After the Linear testchart was printed the theoretical coverage values were replaced with the
effective coverage values with inverting the Murray-Davies spectral print model. The obtained
effective colorant values confirms the dot gain effect. This can be seen in fig. 20.
The modeling process was the same in steps as it was for the Oce ColorWave600 printer.
Once when the colorant coverage was established from actual printed values the estimation of
the color ramps was started. This needed to be conducted for each of the ramp of the linear
testchart separately since the effective coverage is specific for the each colorant channel. The

21
 Spectral Print Reproduction - Modeling and Feasibility

Measured Xerox Cyan reflectances − 0 to 100% coverage Xerox MD Estimated Cyan reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 16: Xerox - measured and estimated spectral reflectances of cyan ramp.

Measured Xerox Magenta reflectances − 0 to 100% coverage Xerox MD Estimated Magenta reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 17: Xerox - measured and estimated spectral reflectances of magenta ramp.

Measured Xerox Yellow reflectances − 0 to 100% coverage Xerox MD Estimated Yellow reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 18: Xerox - measured and estimated spectral reflectances of yellow ramp.

22
 Spectral Print Reproduction - Modeling and Feasibility

Measured Xerox Black reflectances − 0 to 100% coverage Xerox MD Estimated Black reflectances − 0 to 100% coverage
1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6

0.5 0.5

0.4 0.4

0.3 0.3

0.2 0.2

0.1 0.1

0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
λ [nm] λ [nm]

Figure 19: Xerox - measured and estimated spectral reflectances of black ramp.

Xerox Cyan Dot Gain Xerox Magenta Dot Gain

1 1
Effective coverage

Effective coverage

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
Theoretical coverage Theoretical coverage
Xerox Yellow Dot Gain Xerox Black Dot Gain
1 1
Effective coverage
Effective coverage

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 0.5 1 0 0.5 1
Theoretical coverage Theoretical coverage

Figure 20: Effective colorant coverage of Xerox Phaser 6250 printer.

fact that the ramps were monochromatic resulted with using Murray-Davies model in process of
estimation. Predicted spectral reflectances are shown in fig. 16 for cyan, fig. 17 for magenta, fig.
18 and fig. 19 for yellow and black ramp. It can be seen that the estimated reflectances are more
uniform comparing to measured spectral reflectances. In case when the effective coverage was
not considered the uniformity of the spectral reflectances curve is even more expressed because
of neglecting the effect of the dot gain effect and not considering the n value. That will be further
done with YNSN model.

23
 Spectral Print Reproduction - Modeling and Feasibility

4 Spectral printer modeling - results

4.1 Kubelka-Munk theory

In the process of estimating reflectances the Kubelka-Munk equation was employed:
)
K(λ) K(λ) K(λ) ( 1
^r(λ) = 1 + − [( 2+2 ] ) (4.1)
S(λ) S(λ) S(λ) 2

       
K(λ) K(λ) K(λ) K(λ)
=a +b + ... +
S(λ) mixture S(λ) colorant1 S(λ) colorant2 S(λ) paper

s
 2
K(λ) K(λ) K(λ)
R∞ (λ) = 1 + − +2
S(λ) S(λ) S(λ)

This way the relationship between spectral reflectances of the patch and its absorption and scat-
tering characteristics of the patch were established [42] [43]. This was very useful because ratio
K/S of the mixture is given by the sum of the K/S values of the individual colorants which are
primaries of printer [44] [45]. Finally, K/S ratio for all of primaries colorants and their combina-
tions is then used in obtaining the estimated reflectances of the primaries and their combinations
respectively. This is shown in figure 21 for the Oce ColorWave 600 printer and then with figure
22 for the Xerox Phaser 6250.

4.2 Spectral Neugebauer model

Testing this model was important step because it considers all the channels of a printer and by
extending initial Murray-Davies model [46]. This way we were able to calculate spectral re-
flectance value of possible combinations of colorants by taking into account 100% (full) colorant
coverage of primaries. By using equation 2.5 tested patches the following results were obtained
in 23 and 24.
Comparing the results obtained for the modeling of both printers with the Spectral Neuge-
bauer model lead to several conclusions to be read from the tables presented. If comparing
technologies for this particular model the laser technology resulted with more accurate values.
Futher, moving to measured, estimated and mixed combination of the Neugebauer primaries
tested on 108 extracted patches from testchart resulted with the best calculations for the mixed
combination. That value was obtained with the ∆E00.

24
 Spectral Print Reproduction - Modeling and Feasibility

Oce − estimated NP
1
Oce − measured NP
1
0.9
0.9
0.8
0.8
0.7
0.7
0.6

Reflectance
0.6
Reflectance

0.5
0.5

0.4
0.4

0.3 Paper
0.3
Paper
Cyan
0.2 Cyan 0.2 Magenta
Magenta Yellow
Yellow 0.1 Black
0.1
Black
Comb Comb
0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
Wavelength Wavelength

Figure 21: Visualization of measured Neugebauer primaries and Neugebauer primaries estimated using
Kubelka Munk equation for Oce ColorWave 600 printer.

Xerox − measured NP Xerox − estimated NP

1 1

0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6
Reflectance

Reflectance

0.5 0.5

0.4 0.4

0.3 Paper 0.3 Paper

Cyan Cyan
0.2 Magenta 0.2 Magenta
Yellow Yellow
0.1 Black 0.1 Black
Comb Comb
0 0
350 400 450 500 550 600 650 700 750 350 400 450 500 550 600 650 700 750
Wavelength Wavelength

Figure 22: Visualization of measured Neugebauer primaries and Neugebauer primaries estimated using
Kubelka Munk equation for Xerox Phaser 6250 printer.

SN - Océ ColorWave 600

NPs mNPs eNPs mixNPs
mean max std mean max std mean max std
sRMSE 0,059 0,061 0,019 0,062 0,069 0,007 0,061 0,072 0,009
GFC 0,993 0,974 0,008 0,992 0,987 0,004 0,994 0,985 0,005
ΔE 7,98 20,41 2,47 7,82 12,45 2,84 7,02 13,24 2,14
ΔE94 5,8 15,41 2,72 6,2 14,57 3,01 5,96 13,24 3,45
ΔE00 5,2 13,21 1,95 5,97 14,87 2,32 5,48 9,97 2,41

Figure 23: Results of Spectral Neugebauer model for Oce ColorWave 600 printer.

25
 Spectral Print Reproduction - Modeling and Feasibility

SN - Xerox Phaser 6250

NPs mNPs eNPs mixNPs
mean max std mean max std mean max std
sRMSE 0,045 0,051 0,021 0,043 0,132 0,019 0,041 0,112 0,012
GFC 0,996 0,972 0,003 0,995 0,919 0,008 0,996 0,909 0,009
ΔE 7,02 19,01 4,01 7,3 20,47 4,02 6,99 15,45 5,14
ΔE94 5,39 16,21 2,54 5,84 12,47 2,41 5,58 2,24 3,21
ΔE00 5,04 14,95 2,75 5,5 16,81 2,13 5,02 2,04 2,45

Figure 24: Results of Spectral Neugebauer model for Xerox Phaser 6250 printer.

4.3 Yule-Nielsen Modified Spectral Neugebauer Model

With this model the nonlinear relationship between the reflection spectra of paper and colorant
placed was tested. This was done by applying a power function, whose exponent n is fitted
according to a limited set of measured patch reflection spectra. With the non-linear relationship
being applied the results obtained were more accurate than the ones obtained with spectral
Neugebauer model. That can be read, for both of the printers, from 25 and 26

YNSN - Océ ColorWave 600

NPs mNPs eNPs mixNPs
n mean max std n mean max std n mean max std
sRMSE 3,5 0,042 0,049 0,005 3,5 0,047 0,051 0,005 4,3 0,055 0,068 0,004
GFC 4,2 0,994 0,991 0,009 4,5 0,994 0,991 0,002 4,2 0,997 0,984 0,005
ΔE 1,9 7,1 15,32 2,31 2,1 7,5 10,21 1,95 1,9 6,95 11,54 1,95
ΔE94 2,1 5,2 16,14 3,12 2,2 5,96 12,41 2,31 2,1 4,97 10,24 3,2
ΔE00 2,4 4,64 12,98 3,1 2,4 5,02 12,23 2,41 2,7 5,02 8,41 2,1

Figure 25: Results of Yule-Nielsen Modified Spectral Neugebauer model for Oce ColorWave 600 printer.

YNSN - Xerox Phaser 6250

NPs mNPs eNPs mixNPs
n mean max std n mean max std n mean max std
sRMSE 3,3 0,036 0,039 0,009 3,3 0,039 0,091 0,009 3,3 0,049 0,102 0,004
GFC 4,1 0,997 0,978 0,01 3,9 0,995 0,923 0,004 4,6 0,996 0,959 0,007
ΔE 2,3 6,2 11,45 2,14 2,3 6,9 15,41 2,31 2,3 7,1 14,21 3,89
ΔE94 2,3 4,5 10,12 2,01 2,3 4,8 10,21 3,1 2,7 5,21 10,25 2,14
ΔE00 2,0 4,2 9,21 1,95 2,1 4,51 14,32 2,03 2,6 4,91 9,87 1,54

Figure 26: Results of Yule-Nielsen Modified Spectral Neugebauer model for Xerox Phaser 6250 printer.

The calculations obtained for the modeling of both printers with the Yule-Nielsen Modified
Spectral Neugebauer Model resulted with conclusions to be read from the tables presented. When

26
 Spectral Print Reproduction - Modeling and Feasibility

comparing technologies for this particular model the laser technology again resulted with more
accurate values. Comparing measured, estimated and mixed combination of the Neugebauer pri-
maries tested on 108 extracted patches from testchart resulted with the best calculations for the
patches not being estimated or mixed but measured. That value was obtained with the ∆E00.
Introducing optimal n value in this model has shown results being noticeable improved. This is
because now not only mechanical but the optical dot gain was considered.

4.4 Cellular spectral Neugebauer Model

With testing cellular spectral Neugebauer model the goal was to reduce the space over which
interpolation is performed. This was done by providing more primaries by measuring each col-
orant at 0, 50, and 100% area coverage. This also means that the printer gamut is subdivided into
smaller cells comparing to binary system of placing colorant (placed: 100% or not placed:0%) on
substrate. By applying all these constrains the following results were obtained for Oce ColorWave
600 27 and Xerox Phaser 6250 28 In case of the calculations obtained for the modeling of both

cSN - Océ ColorWave 600

NPs mNPs eNPs mixNPs
mean max std mean max std mean max std
sRMSE 0,04 0,049 0,018 0,041 0,062 0,009 0,042 0,059 0,012
GFC 0,997 0,98 0,005 0,997 0,967 0,012 0,998 0,98 0,004
ΔE 6,5 10,45 1,95 6,8 12,45 2,14 6,52 13,54 2,45
ΔE94 5,1 12,45 2,14 5,45 13,21 1,92 5,12 12,14 4,21
ΔE00 4,1 10,27 2,12 4,9 13,71 2,52 4,15 11,25 2,41

Figure 27: Results of Cellular spectral Neugebauer model for Oce ColorWave 600 printer.

cSN - Xerox Phaser 6250

NPs mNPs eNPs mixNPs
mean max std mean max std mean max std
sRMSE 0,03 0,041 0,004 0,032 0,049 0,011 0,03 0,051 0,012
GFC 0,997 0,991 0,004 0,997 0,991 0,002 0,998 0,987 0,004
ΔE 5,98 13,21 2,45 6,12 12,54 2,51 6,1 15,12 2,98
ΔE94 4,15 13,25 2,02 4,2 11,49 1,97 4,5 14,21 3,14
ΔE00 3,95 11,47 1,98 4,1 12,03 1,48 4,23 15,21 3,87

Figure 28: Results of Cellular spectral Neugebauer model for Xerox Phaser 6250 printer.

printers with the Cellular spectral Neugebauer Model resulted with conclusions to be read from

27
 Spectral Print Reproduction - Modeling and Feasibility

the tables presented. When comparing technologies for this particular model the laser technology
again resulted with more accurate values. Comparing measured, estimated and mixed combina-
tion of the Neugebauer primaries tested on 108 extracted patches from testchart resulted with
the best calculations for the patches that were measured have shown the best result.Nevertheless,
values very close to that are the ones obtained with mix both measured and estimated vales of
Neugebuer Primaries. That most accurate value was obtained with the ∆E00. Introducing the
smaller cell (range 0,0.5,1 instead only 0 to 1) in this model has shown results being signifi-
cantly improved. This is due to decreasing the size of the cell that enables interpolation to be
more precise.

4.5 Cellular Yule-Nielsen spectral Neugebauer model

When the testing of cellular spectral Neugebauer model was extended with addition of a Yule-
Nielsen factor the accuracy was shown to be improved. The reason was that with this model(cYNSN),
unlike with plain YNSN model, physical printer transfer function is sampled on more than the 2k
Neugebauer primaries for a k colorant printer. That was the 50% coloranat coverage added that
made cell smaller, therefore more accurate. The results can be read from 29 for Oce ColorWave
600 and 30 for Xerox Phaser 6250 printer. The calculations obtained for the modeling of both

cYNSN - Océ ColorWave 600

NPs mNPs eNPs mixNPs
n mean max std n mean max std n mean max std
sRMSE 3,1 0,031 0,04 0,009 3,4 0,035 0,039 0,004 3,5 0,035 0,045 0,009
GFC 3,9 0,997 0,988 0,007 4,2 0,997 0,995 0,002 4,1 0,998 0,991 0,006
ΔE 1,5 5,45 14,31 2,2 1,6 5,9 9,52 2,12 1,9 6,15 10,25 2,35
ΔE94 1,8 3,82 12,38 3,4 2,1 5,1 9,89 2,98 2,1 4,25 10,89 3,98
ΔE00 2,1 3,5 10,6 2,9 2,4 3,8 10,25 2,54 2,3 4,83 9,45 3,25

Figure 29: Results of Cellular Yule-Nielsen spectral Neugebauer model for Oce ColorWave 600 printer.

cYNSN - Xerox Phaser 6250

NPs mNPs eNPs mixNPs
n mean max std n mean max std n mean max std
sRMSE 3,3 0,028 0,031 0,008 3,2 0,032 0,038 0,007 4,5 0,035 0,055 0,004
GFC 4,1 0,998 0,989 0,009 4,1 0,997 0,988 0,007 5,2 0,997 0,987 0,006
ΔE 2,3 5,15 12,35 3,45 2,4 6,12 14,58 3,65 2,3 6,52 13,24 3,47
ΔE94 2,3 3,98 10,78 2,45 2,9 4,2 11,24 2,54 2,3 4,52 9,66 2,1
ΔE00 2,0 3,2 8,52 2,86 2,7 4,1 13,22 2,02 2,0 4,98 8,41 2,33

Figure 30: Results of Cellular Yule-Nielsen spectral Neugebauer model for Xerox Phaser 6250 printer.

printers with the Cellular Yule-Nielsen spectral Neugebauer model resulted with conclusions to
be read from the tables presented. When comparing technologies for this particular model the
laser technology does not show more accurate values comparing to Oce printer at it was pre-
vious case. When used measured spectral reflectances of the Neugebauer primaries Xerox does

28
 Spectral Print Reproduction - Modeling and Feasibility

give the higher accuracy values obtained. Neverthless, in estimated and mixed Neugebeuer pri-
maries Oce printer results with more accurate values. this is in case of calculation obtained with
the ∆E00. Introducing optimal n value and making the cell smaller (range 0,0.5,1 instead only
0 to 1) in this model has shown results being significantly improved as well as having optimal
nvalues considered. This is result of to decreasing the size of the cell that enables interpolation to
be more precise, as well as taking into account not only mechanical but the optical dot gain effect.

29
 Spectral Print Reproduction - Modeling and Feasibility

5 Feasibility of spectral print reproduction

In previous chapters spectral printer modeling was explained. Spectral printer models were also
performed for two printers of different printing technology. The results were different for laser
technology comparing to crystal point which is extension of inkjet technology. Both of the printers
were using four channels for spectral print reproduction. This fact put in front of us interesting
question: will increasing the number of colorant channels also increase possibility to reach the
exact spectral color reproduction. In order to answer this question fundamental knowledge of
halftoning process should be considered [47] [48] [49]. That process can be started by placing
two colorants on paper substrate and then comparing the obtained complete subset of colors
with the same result of placing four colorants. From that example it is possible to observe that
color gamut, which is obtained with complete subset of colors printed, plays very important role
in making any further conclusion. The subset of colors which can be accurately represented in a
given circumstance, such as within a given color space or by a certain output device.The process
of printing any combinations of colorants nowadays is set to high accuracy that is followed by
high level reproduction. This is due to fast development of new technologies that enable better
control of the printer and printing system [50]. Starting with the spectral color reproduction
input instruments give high precision of data collected but the challenge is still in creating the
precise communication between the input measurements and wanted printed material. Starting
from the point where the issues of creating the output product are well known the importance
of assuring the spectral print of that product is possible at all becomes more significant and
valuable information. The reason is the waste of resources, colorants and paper as well as time.
It is important to have confirmation that the spectral reflectances wanted to be reproduced are
possible with the given characteristic of printer available. For this reason in this part of thesis
work the process of deciding weather the spectral print reproduction is possible or not was tried
to be automated. The algorithm consisted of few steps that lead to final yes or no was proposed.
Having the strong twelve colorant printer on disposition, as well as matching set of different
paper substrates available, made this process to give results from which in the end the valuable
conclusion on feasibility was made.

5.1 Collaboration with retailer

In this part of thesis focus was on feasibility of particular product. The collaboration with retail
store resulted in obtaining the real samples and testing the possibility of spectral print reproduc-
tion with printer in our lab. The challenge was in trying to reduce the cost that retailer currently
incur on one front of the retail color development process. That particularly involves recurring
expenditure on 5 cm x 5 cm fabric samples that are produced to within 0.8 DE CMC. This is valid
for illuminant UL3000 and 10 degrees observer. Designers use these physical representations
when they are deciding which shade an item of garment product will be, or which shade needs
to be placed near another in a printed or yarn dyed pattern on a garment product. The existing

30
 Spectral Print Reproduction - Modeling and Feasibility

set of roughly 4000 colors is supporting the design process within the color set that designers are
able to pick from. Once their colors for a season and brand are picked, 5cm x 5cm fabric samples
are ordered to work with. The 4000 colors are supported using spectral data. That reference
helps keep the color on shade through many production runs, i.e. of color chips, of the color
standards, and of the garments that are specified to be in close tolerance of these colors. With
that given one set of spectral data for all colors, within scope of these thesis feasibility of printing
particular colors out on as needed basis was tested. The possibility of printing these color sam-
ples versus purchasing them from a dyeing company would significantly save in cost per color.
That is why the important step was focus on investigation the feasibility of being able to produce
these colors accurately and consistently over the many variables that arise during consecutive
runs. The starting issue in getting the print head create the wanted output, was to replace the
D50 table with a UL3000 illuminant. The reason for that was the fact that color standards are
made using reactive dyes, and the printers are limited to a significantly smaller set of colorants.
Therefore, from a reflectance curve matching standpoint it was expected that there will likely
be gaps. To overcome that problem the try to match to UL3000 illuminant was very important.
Another challenge that was related to gamut is the difference in available chroma in the chro-
mophores. Given that there are spectral data for all 4000 colors available, the brightest colors to
be produced to produce are known as well but only the right combination of paper and colorants
will create the maximum chroma. This is due to having the spectral data for the targets it would
be easy to generate the L*,a*,b*, C, and h values for all colors. From this step then it is possible
to determine the maximum chroma needed. This would be a base ground for possible realization
of coverage for given gamut [51] [52].

5.2 Color Gamut

In color theory, the gamut of a device or process is that portion of the color space that can be
represented, or reproduced. Generally, the color gamut is specified in the hue-saturation plane,
as a system can usually produce colors over a wide intensity range within its color gamut. For a
subtractive color system, as used in printing, the range of intensity available in the system is for
the most part meaningless without considering system-specific properties. That property is the
illumination of the colorant being placed on paper [53] [54] [55]. In cases when certain colors
cannot be expressed within a particular color model, those colors are said to be out of gamut.
For example, while pure red can be expressed in the RGB color space, it cannot be expressed in
the CMYK color space; pure red is out of gamut in the CMYK color space. Even though modern
techniques allow increasingly good approximations, at this moment device that is able to repro-
duce the entire visible color space is still an unrealized goal within the engineering of printing
processes. In processing a digital image the RGB color space is the most convenient color model
used. Printing the image requires transforming the image from the original RGB color space to
the printer’s CMYK color space. During this process, the colors from the RGB which are out of
gamut must be converted to approximate values within the CMYK space gamut. Simply trimming
only the colors which are out of gamut to the closest colors in the destination space would burn
the image. There are several algorithms approximating this transformation, but none of them is
perfect. The reason of that is that those colors are simply out of the target device’s capabilities.

31
 Spectral Print Reproduction - Modeling and Feasibility

This is why identifying the colors in an image which are out of gamut in the target color space
as soon as possible during processing is critical for the quality of the final print reproduction
product.

5.3 Problem specification

What was done by now was the reproduction of the same set of 4000 colors using pigments in an
injection molded process. Within that the knowledge that reflectance of colors made while using
reactive dyes helped in match and use reflectance curves of pigments. That was why the whole
process wanted to be moved to one step further - ink and paper combination. The samples used
for our calculation were subset of original set consisted of 4000 samples. The subset consisted
of 32 data samples(details in Appendix). Spectral data of a textile standard and a plastic stan-
dard for the 10 degree observer weighted illuminant were given. The colors made by injection
molding were used as color standards which suppliers in the hard goods area need to match.
The idea was to have ’pre-translated’ the color difference between textile dyes used in Apparel
and Soft home products, and pigments used in Hard home areas.The main goal was to have the
L*a*b* values of printed samples to match the L*a*b* values of the standard or have the color
difference in terms of CMC DE in UL3000, which is store light, for 10 degree observer between
the two be as low as possible. An acceptable difference limit was defined with value of is 0.8 DE
CMC. Those that fall outside of that difference up to 1.2 can be given special consideration by
using visual determination of color difference. Those above 1.2 can be identified as out of gamut
for the system under investigation. Spectral reflection data was obtained on spectrophotometer
Datacolor SF600 Plus CT for each associated color. The values of reflectance were measured by
this instrument at 10 nm interval starting at 400nm. Datacolor SF600 plus CT has an integrating
sphere for illumination of the sample and the detector for the spectral analyzer is placed at 8
degrees from normal to the surface of the aperture plane (sample port). Sample ports are config-
urable depending on size of sample being presented. LAV (Large Area View) aperture plate is 30
mm illuminated and 26nm measured. This is important for ability to maximize the surface area
to be measured which is ideal to measure color without regard to a textured surface.The instru-
ment is dual-beam spectrophotometer with diffuse illumination and high black trap performance
[56] [44].
The metric that defines a difference measure used here is CMC. It was created in 1984. by the
Colour Measurement Committee of the Society of Dyers and Colourists [57]. It is based on the
L*C*h color model. Named after the developing committee, their metric is called CMC l:c. It has
two parameters: lightness (l) and chroma (c), allowing the users to weight the difference based
on the ratio of l:c that is deemed appropriate for the application. Commonly-used values are 2:1
for acceptability and 1:1 for the threshold of imperceptibility. It is defined with the equations
5.1, 5.2 and 5.2. The distance of a color (L∗2 , C∗2 , h2 ) to a reference (L∗1 , C∗1 , h1 ) is as follows:
s
 ∗ 2  ∗ 2  2
L2 − L∗1 C2 − C∗1 ∆H∗ab
∆E∗CMC = + + (5.1)
lSL cSC SH

32
 Spectral Print Reproduction - Modeling and Feasibility

where

0.511 L∗1 < 16 0.0638C∗1
SL = 0.040975L∗ SC = + 0.638 SH = SC (FT + 1 − F) (5.2)
1
1+0.01765L∗ L∗1 ≥ 16 1 + 0.0131C∗1
1

and where
s 
0.56 + |0.2 cos(h1 + 168◦ )| 164◦ ≤ h1 ≤ 345◦
4
C∗1
F= T= (5.3)
0.36 + |0.4 cos(h1 + 35◦ )| otherwise
∗4
C1 + 1900

Figure 31: Datacolor SF600 spectrophotometer.

Figure 32: Lab values of retailer’s color samples.

5.4 ICC3D
With given constraints of the error allowed within delta CMC value 0.8 the process of spectral
color reproduction was decided to be started with checking the feasibility. Two printers that were
used for spectral print modeling were considered to be employed, Xerox Phaser 6250 color laser
and Oce ColorWave 600 printer. The process was started with numerical and visual representa-
tion of their gamuts. This was done by using color ramps of separated printed colorants. Next,
spectral reflection data of ramp was measured with and finally represented with the L*a*b* val-
ues by means ICC3D software. The functionality of the ICC3D software is the ability to read
color data as an image in some standard format or that are extracted by the gamut tag of an
ICC profile. It also enables the color data to be displayed in different ways in various 3D color
spaces [58]. This is very useful for visualizing the whole process of checking the spectral print
feasibility. The L*a*b* values of the four channels printer and their gamuts are shown with fig.
33 for Xerox Phaser 6250 and fig. 33 for the Oce ColorWave 600 printer.

33
 Spectral Print Reproduction - Modeling and Feasibility

Figure 33: 3D gamut visualization of Xerox Phaser 6250 printer is shown on the right side. On the left side
L*a*b* values of ramp printed separately for all four channels is presented.

Figure 34: 3D gamut visualization of Oce ColorWave 600 printer is shown on the right side. On the left side
L*a*b* values of ramp printed separately for all four channels is presented.

The printer that was mentioned in previous chapters, twelve channel HP DesignJet Z3200,
was now employed. The reason that it was not used in modeling process was time that it took us
to get the corresponding RIP. Nevertheless, in this part of thesis it was crucial element because of
its much better gamut size comparing to the four channel printers. The gamut of the HP Design-
Jet Z3200 printer was obtained for five different papers and visualized using ICC3D. Calibration
process of the HP DesignJet Z3200 printer results with a printed color chart made of a ramps for
each channel. These ramps were then measured manually with Eye One Pro spectrophotometer
because of the specific hexagonal shape of color patches. Linear testchart was also measured with
the same spectrophotometer and then spectral reflectances obtained with it were compared with

34
 Spectral Print Reproduction - Modeling and Feasibility

the ones obtained with Spectrolino. The RMSE for these two measurements was then calculated
to be 0.001. For this reason representing and comparing gamut values for four colorant printers
with HP DesignJet Z3200 was continued to be done with certainty of precision [59] [59].

5.5 Proposed Algorithm

With the change of the paper substrate the diversity of possible spectral print reproduction was
goal to be checked. The fact that in process of reaching final confirmation or negation of spectral
print possibility many different steps are conducted it was very important to make that process
decreased in time consumed [60] [61] [58]. In this particular case there were 32 spectral data
color samples given for which feasibility was to be checked. Nevertheless, as mentioned before,
that was just the subset of initial amount wanted to be reproduced in form of spectral print. For
that reason the algorithm that enables the whole process, from spectral reflectance input to final
decision of sending to printer or not, automated was proposed.
At the beginning of the proposed algorithm the input data required was the spectral reflectance
data measurements. It needs to be noted here that it is important take care of the geometry of
measurements for the main goal. When (if) spectral print reproduction of wanted object was
confirmed to be feasible the final print reproduction should be measured under same conditions,
geometry and illuminats. Since the next step was mapping the color gamut of a initial values
to the gamut given by output mechanism, printer, the values were represented with an image.
For the image format Tagged Image File Format was chosen. TIFF image format enables data
obtained from spectral reflectances measured under certain illuminant to be presented separately
for each of L*, a* and b* channels. If this wanted to be replaced with CMYK values TIFF is
again the best choice because in that case it represents four channels separately. By use of TIFF
format the conversion to RGB space was skipped, regarding important use of displays, during
the whole process. This helps in making the whole process faster and simple. After the TIFF
image was created in CIELAB color space next step was to map that image file, which is the
representation of the spectral reflectance of the object wanted to be printed, to the gamut of a
printed for a certain paper. In process of mapping ICC3D software was employed and constrains
set was minimum delta and constant hue in order to achieve the best perceptual match of final
reproduction. Obtained mapped image is again in TIFF format [62] which enables to extract
the L*, a* and b* channel data. Next, that data was used in step of comparing the differences
between the initial TIFF image and mapped TIFF image. Then the calculation of the differences
for several color spaces in case of several paper substrates under different illuminants was done.
Visualization of the points that are out of gamut for a certain paper was presented in order to
be able to notice which part of gamut should be considered to be extended if possible was the
next step done. Finally, with the constraints to be considered, i.e. value more than 1.2 for delta
CMC overcomes the aim, the spectral color points were confirmed or not to be feasible to print
within the gamut possibilities for specific combination of colorants/paper of printer given. The
feasibility of the defined input data from retailer was then tested with this proposed method for
five different papers. The scheme can be seen in fig. 35.

35
 Spectral Print Reproduction - Modeling and Feasibility

OBJECT
Illuminant Geometry

Spectral reflectance data

[Spectrophotometer] TIFF image Spectral reflectance data
[Spectrophotometer]

Mapping
[ICC3D] Substrate gamut
Constraints
[min Delta, max hue...]

Ini!al TIFF vs. mapped TIFF

Calculated Difference

Feasible
YES NO

Figure 35: Feasibility of spectral print reproduction algorithm scheme.

5.6 HP Heavy Coated paper

HP Heavy Coated paper

A ΔE Δ94 ΔCMC ΔE00
min max avg min max avg min max avg min max avg
D50 0,21 11,38 2,41 0,15 10,56 1,98 0,21 18,82 2,66 0,17 7,65 1,57
D65 0,21 11,02 2,48 0,15 10,17 2,02 0,21 18,12 2,68 0,17 7,37 1,60
A 0,26 11,02 2,45 0,11 10,49 1,99 0,13 18,77 2,80 0,10 7,54 1,58
C 0,26 11,02 2,50 0,20 12,96 2,83 0,13 18,77 2,85 0,10 7,54 1,62
F2 0,21 16,13 4,02 0,20 12,96 2,83 0,28 16,67 3,35 0,21 8,84 2,30
UL3000 0,21 16,13 4,07 0,20 12,96 2,87 0,28 17,37 3,42 0,21 8,84 2,33

Figure 36: Results of gamut mapping to paper A given with different color differences equations under 6
illuminants.

The paper that the testing the feasibility was started with was HP Heavy Coated paper. That
paper is 130g/m2 paper often used while high quality print reproduction is to be obtained. From
the fig. 36 the calculations can be read. The important results form the table given are the one
for CMC color difference under UL3000 illuminant. This is the determination point of spectral
print feasibility. As stated in problem given the spectral print should match under illuminant
UL3000 but the rest illuminants match is highly desirable. For that reason the other illuminants
(D50,D65,A,C,F2) were taken into account. In this particular case the input spectral reflectance
data come from the cloth object which is the reason of delta CMC being important. Nevertheless,
a few other color differences were employed as well to see how the resulting calculations behave
in different mathematical color environment. This is also due to the fact that the algorithm can

36
 Spectral Print Reproduction - Modeling and Feasibility

be used not only for spectral data obtained from measuring the garnament but the variety of
objects and products. In this particular case of paper it can be seen that the minimum CMC
difference is obtained under illuminnat different than UL3000 but even for the aimed illuminant
the results are satisfied. This is in case of minimum but if the average is to be read, the 1.8 and
0.8 threshold values become overpassed. This can be misleading number because the significant
amount of samples is still above 1.2 values and 0.8. This can be seen from the histogram in
fig. 37. From there is easier to observe the amount of samples that are feasible to be produced in
spectral print reproduction, in particular 13 samples are under CMC 0.8 value and 6 under CMC
1.2 value of total 32 samples.

13 13
14
12
10
8 6
6
4
2
0
0 - 0.8 0.81 - 1.2 1.2 <

32 Color Samples

Figure 37: Histogram of resulting CMC difference for 32 color samples mapped to printer gamut on paper
A.

In observing the values that go over the 1.2 CMC value, the particular color points that are
the most far away from wanted values to be mapped are in yellow and dark blue(black) part of
the color gamut that can be seen from the fig. 38. This is the crucial information where to look
for gamut extension in order to achieve the wanted spectral reproduction.

Figure 38: Visualization of initial color values and gamut of print created with paper A before mapping, in
3D space.

37
 Spectral Print Reproduction - Modeling and Feasibility

In the end, starting from the intimal 3 separated channels in TIFF format the mapped version
to gamut obtained on HP Heavy Coated paper is created and presented visually with L*a*b*
values in color ramp of target colors with the fig. 39

Figure 39: Color samples mapped to gamut of HP Heavy Coated paper under illuminant UL3000 for 10
degree observer.

5.7 HP Professional Matte Canvas

The HP Professional Matte Canvas is 430g/m2 paper often for fine art reproduction. This paper
was chosen for the texture that resembles more to the cloth that the HP Heavy Coated paper. In
this particular case of paper it can be seen from the fig. 40 that the maximum CMC difference
is obtained under illuminat different than UL3000 but even for the aimed illuminant the results
are satisfied. This is in case of minimum but if the average is to be read, the 1.8 and 0.8 thresh-
old values become overpassed. This is again, as mentioned before, misleading number because
the significant amount of samples is still above 1.2 values and 0.8. This can be seen from the
histogram in fig. 41. From there is easier to observe the amount of samples that are feasible to be
produced in spectral print reproduction, in particular 20 samples are under CMC 0.8 value and
5 under CMC 1.2 value of total 32 samples. In observing the values that go over the 1.2 CMC

HP Professional Ma!e Canvas

B ΔE Δ94 ΔCMC ΔE00
min max avg min max avg min max avg min max avg
D50 0,21 11,02 1,96 0,11 10,49 1,63 0,13 18,77 2,50 0,10 7,54 1,28
D65 0,21 10,65 2,01 0,11 10,11 1,66 0,13 18,07 2,51 0,10 7,25 1,31
A 0,21 11,02 2,41 0,11 10,49 1,91 0,13 18,77 2,68 0,10 7,54 1,54
C 0,21 10,65 1,98 0,11 10,11 1,64 0,13 18,07 2,49 0,10 7,25 1,30
F2 0,21 11,25 3,11 0,20 9,69 2,41 0,28 16,67 3,11 0,21 8,02 1,96
UL3000 0,21 10,28 2,04 0,11 9,72 1,69 0,13 17,37 2,52 0,10 6,96 1,33

Figure 40: Results of gamut mapping to paper B given with different color differences equations under 6
illuminants.

value, the particular color points that are the most far away from wanted values to be mapped
are in yellow and dark blue(black) part of the color gamut that can be seen from the fig. 42.
This is the crucial information where to look for gamut extension in order to achieve the wanted
spectral reproduction.
Again, starting from the intimal 3 separated channels in TIFF format the mapped version to
gamut obtained on HP Professional Matte Canvas is created and presented visually with L*a*b*
values in color ramp of target colors with the fig. 43

5.8 HP Artist Matte Canvas

The HP Artist Matte Canvas is 380g/m2 paper often for fine art reproduction. This paper was
chosen for the texture that resembles even more to the cloth that the HP Professional Matte
Canvas. In this particular case of paper it can be seen from the fig. 44 that the maximum CMC

38
 Spectral Print Reproduction - Modeling and Feasibility

20
20

15

10 7
5
5

0
0 - 0.8 0.81 - 1.2 1.2 <

32 Color Samples

Figure 41: Histogram of resulting CMC difference for 32 color samples mapped to printer gamut on paper
B.

Figure 42: Visualization of initial color values and gamut of print created with paper B before mapping, in
3D space.

Figure 43: Color samples mapped to gamut of HP Professional Matte Canvas under illuminant UL3000 for
10 degree observer.

difference is obtained under illuminnat UL3000 so in comparison with the rest illuminants the
calculations obtained are satisfied. This is in case of minimum but if the average is to be read, the
1.8 and 0.8 threshold values become overpassed. This is again, as mentioned before, misleading
number because the significant amount of samples is still above 1.2 values and 0.8. This can be
seen from the histogram in fig. 45. From there is easier to observe the amount of samples that
are feasible to be produced in spectral print reproduction, in particular 20 samples are under
CMC 0.8 value and 5 under CMC 1.2 value of total 32 samples. In observing the values that
go over the 1.2 CMC value, the particular color points that are the most far away from wanted
values to be mapped are in yellow and dark blue(black) part of the color gamut that can be seen

39
 Spectral Print Reproduction - Modeling and Feasibility

HP Ar!st Ma"e Canvas

C ΔE Δ94 ΔCMC ΔE00
min max avg min max avg min max avg min max avg
D50 0,21 11,02 2,45 0,11 10,49 1,96 0,13 18,77 2,73 0,10 7,54 1,57
D65 0,21 10,65 1,91 0,11 10,11 1,56 0,13 18,07 2,39 0,13 7,25 1,26
A 0,21 11,02 2,47 0,11 10,49 1,97 0,13 17,05 2,73 0,10 7,54 1,58
C 0,21 11,04 2,47 0,11 10,50 1,97 0,13 17,77 2,73 0,10 7,50 1,58
F2 0,21 11,02 3,31 0,20 10,49 2,50 0,28 18,77 3,19 0,21 7,54 2,02
UL3000 0,21 10,28 1,93 0,11 9,72 1,57 0,13 17,37 2,36 0,10 6,96 1,27

Figure 44: Results of gamut mapping to paper C given with different color differences equations under 6
illuminants.

20
20

15

10 7
5
5

0
0 - 0.8 0.81 - 1.2 1.2 <

32 Color Samples

Figure 45: Histogram of resulting CMC difference for 32 color samples mapped to printer gamut on paper
C.

from the fig. 46. From here the lead to direction to look for gamut extension in order to achieve
the wanted spectral reproduction is created.

Figure 46: Visualization of initial color values and gamut of print created with paper C before mapping, in
3D space.

Again, starting from the intimal 3 separated channels in TIFF format the mapped version to

40
 Spectral Print Reproduction - Modeling and Feasibility

gamut obtained on HP Artist Matte Canvas is created and presented visually with L*a*b* values
in color ramp of target colors with the fig. 47

Figure 47: Color samples mapped to gamut of HP Artist Matte Canvas under illuminant UL3000 for 10
degree observer.

5.9 HP Universal Coated Paper

The HP Universal Coated Paper is 95g/m2 paper often for universal print reproduction. This pa-
per was chosen for establishing the relation between specific requirements with universal paper
substrate in order to see the behavior in daily print materials.In this particular case of paper it
can be seen from the fig. 48 that the maximum CMC difference is obtained under illuminnat
UL3000 so in comparison with the rest illuminants the calculations obtained are satisfied. This
is in case of minimum but if the average is to be read, the 1.8 and 0.8 threshold values be-
come overpassed. This is again, as mentioned before, misleading number because the significant
amount of samples is still above 1.2 values and 0.8. This can be seen from the histogram in
fig. 49. From there is easier to observe the amount of samples that are feasible to be produced
in spectral print reproduction, in particular 18 samples are under CMC 0.8 value and 6 under
CMC 1.2 value of total 32 samples. In observing the values that go over the 1.2 CMC value,

HP Universal Coated Paper

D ΔE Δ94 ΔCMC ΔE00
min max avg min max avg min max avg min max avg
D50 0,21 12,47 2,41 0,15 11,72 2,02 0,21 20,92 2,82 0,17 8,52 1,60
D65 0,21 12,16 2,42 0,15 11,02 1,97 0,21 19,57 2,52 0,17 8,09 1,59
A 0,21 12,51 3,14 0,20 11,40 2,39 0,28 20,27 2,90 0,21 8,37 1,92
C 0,21 12,16 2,33 0,15 11,02 1,91 0,21 19,57 2,47 0,17 8,09 1,54
F2 0,21 14,79 3,55 0,20 13,26 2,64 0,28 17,42 3,00 0,21 9,27 2,17
UL3000 0,21 11,82 2,24 0,15 10,64 1,82 0,21 18,87 2,36 0,17 7,80 1,47

Figure 48: Results of gamut mapping to paper D given with different color differences equations under 6
illuminants.

the particular color points that are the most far away from wanted values to be mapped are in
yellow and dark blue(black) part of the color gamut that can be seen from the fig. 50. From
here the lead to direction to look for gamut extension in order to achieve the wanted spectral
reproduction is created. Again, starting from the intimal 3 separated channels in TIFF format the
mapped version to gamut obtained on HP Artist Matte Canvas is created and presented visually
with L*a*b* values in color ramp of target colors with the fig. 51

5.10 HP Recycled Bond Paper

The HP Recycled Bond Paper is 80g/m2 paper often for low print reproduction. This paper was
chosen for establishing the realtion between specific requirments with the paper substrate sup-
porting the green technologies.This was to check the results with taking into account the future
possibility of spectral print reproduction on reused paper substrate and in that way making it

41
 Spectral Print Reproduction - Modeling and Feasibility

18
20

15

8
10 6

0
0 - 0.8 0.81 - 1.2 1.2 <

32 Color Samples

Figure 49: Histogram of resulting CMC difference for 32 color samples mapped to printer gamut on paper
D.

Figure 50: Visualization of initial color values and gamut of print created with paper D before mapping, in
3D space.

Figure 51: Color samples mapped to gamut of HP Universal Coated Paper under illuminant UL3000 for 10
degree observer.

more affordable and desirable for environment.In this particular case of paper it can be seen
from the fig. 52 that the maximum CMC difference is obtained under illuminnat UL3000 so in
comparison with the rest illuminants the calculations obtained are satisfied. This is in case of
minimum but if the average is to be read, the 1.8 and 0.8 threshold values become overpassed.
This is again, as mentioned before, misleading number because the significant amount of sam-
ples is still above 1.2 values and 0.8. This can be seen from the histogram in fig. 53. From there
is easier to observe the amount of samples that are feasible to be produced in spectral print re-

42
 Spectral Print Reproduction - Modeling and Feasibility

production, in particular 7 samples are under CMC 0.8 value and 3 under CMC 1.2 value of total
32 samples. From these results it can be concluded that quality of the paper strongly affects the
feasibility of the spectral print reproduction. In observing the values that go over the 1.2 CMC

HP Recycled Bond paper

E ΔE Δ94 ΔCMC ΔE00
min max avg min max avg min max avg min max avg
D50 0,26 22,44 7,82 0,21 19,70 5,23 0,28 34,64 6,74 0,25 14,98 4,30
D65 0,26 22,56 7,86 0,21 19,32 5,17 0,28 33,95 6,66 0,25 14,67 4,25
A 0,26 24,48 8,53 0,21 19,83 5,58 0,28 34,71 7,13 0,25 15,17 4,62
C 0,26 22,11 7,64 0,21 19,32 5,06 0,28 33,95 6,55 0,25 14,67 4,16
F2 0,26 30,78 10,09 0,24 18,41 6,15 0,33 32,47 7,61 0,29 14,07 5,10
UL3000 0,26 22,57 7,57 0,21 18,79 5,06 0,28 33,16 6,54 0,25 14,38 4,16

Figure 52: Results of gamut mapping to paper E given with different color differences equations under 6
illuminants.

22
25

20

15
7
10
3
5

0
0 - 0.8 0.81 - 1.2 1.2 <

32 Color Samples

Figure 53: Histogram of resulting CMC difference for 32 color samples mapped to printer gamut on paper
E.

value, the particular color points that are the most far away from wanted values to be mapped
are in yellow and dark blue(black) part of the color gamut that can be seen from the fig. 54.
From here the lead to direction to look for gamut extension in order to achieve the wanted spec-
tral reproduction is created. Starting from the intimal 3 separated channels in TIFF format the
mapped version to gamut obtained on HP Artist Matte Canvas is created and presented visually
with L*a*b* values in color ramp of target colors with the fig. 55

5.11 Comparing results

After all substrates were tested and the feasibility was given with the results obtained the com-
parison of the five papers was done. This can be seen from the histogram in fig. 56. The fact
that calculating the average difference from the ramp of target set of colors to be spectrally
reproduced is not giving the correct answer need to be noted. For that reason histogram rep-
resentation was used for more thorough conclusions about the feasibility. This move brings the
improvement to visual representation of the data as well as to the calculation results. Finally,

43
 Spectral Print Reproduction - Modeling and Feasibility

Figure 54: Visualization of initial color values and gamut of print created with paper E before mapping, in
3D space.

Figure 55: Color samples mapped to gamut of HP Recycled Bond Paper under illuminant UL3000 for 10
degree observer.

25

20

15 0 - 0.8
0.81 - 1.2
10 1.2 <

0
paper A papaer B paper C paper D paper E

Figure 56: Histogram of resulting CMC difference for initial 32 color samples mapped to printer gamut on
papers A,B,C,D and E.

all of the mapped values are given with the L*a*b* values that are possible to be reproduced.
Observing it carefully gives the opportunity to apply the 3 visualization of gamuts to real samples
wherease the wanted extension of gamut in yellow and blue (black) direction of gamut space is
as well noticed. If focused on the yellow and blue samples of given ramp the deviation from the

44
 Spectral Print Reproduction - Modeling and Feasibility

initial ramp can be seen even with the bare eye. It is somewhat more difficult to noticed yellow
mapped samples difference that blue ones due to ability of human visual system.

Initial
A
B
C
D
E

Figure 57: Color samples mapped to gamut of five different papers under illuminant UL3000 for 10 degree
observer. The arrows point the samples most affected with the gamut mapping. Initial L*a*b* values are
given in top ramp followed by papers A,B,C,D,and E towards the bottom. The arrows point the samples
most affected with the gamut mapping.

45
 Spectral Print Reproduction - Modeling and Feasibility

6 Conclusion

During the work on this master thesis two questions that were initially stated were aimed to
be answered. In order to successfully model the printer for final goal of spectral reproduction
of garnament the Kubelka-Munk theory was employed for estimation of the Neugebauer pri-
maries. This was done in order of saving time and resources. Combination of both measured and
estimated spectral reflectances was also employed in testing the models. Further more, Kubelka-
Munk equation was first time used for testing the cellular extension of spectral Neugebauer
model and Yule-Nielsen spectral Neugebauer model. This resulted in even more accurate values
of models tested. In the future work making the cell even smaller than only 0.5 of the unity
should give even more better and accurate results.
In the second part of the thesis focus was on the feasibility of the spectral print reproduction. For
that reason many different steps were included in one resulting algorithm. With this proposed
method the whole process should be decreased in time for the bigger amount of data as input.
The twelve colorant printer was tested for given target and that resulted with some deviations
from wanted spectral print. This way it was possible to see exact which part of the gamut is
lacking in contribution of the spectral print. The feasibility check was done for the primaries of
the printer and this could be done with checking the overlaps of its primaries by estimating the
secondaries, tertiaries and so on. This is step that is being worked on, yet due to deadline given
the results are not presented here. The main drawback and one of the challenge to improve the
algorithm is making it possible not only for the integer CIELAB values as input.

46
 Spectral Print Reproduction - Modeling and Feasibility

7 APPENDIX A

Results of mapping obtained with paper A (HP Heavy Coated paper) under different illuminats.

Figure 58: Color samples mapped to gamut of HP Heavy Coated paper under illuminant D50 for 10 degree
observer.

Figure 59: Color samples mapped to gamut of HP Heavy Coated paper under illuminant D65 for 10 degree
observer.

Figure 60: Color samples mapped to gamut of HP Heavy Coated paper under illuminant A for 10 degree
observer.

Figure 61: Color samples mapped to gamut of HP Heavy Coated paper under illuminant C for 10 degree
observer.

Figure 62: Color samples mapped to gamut of HP Heavy Coated paper under illuminant F2 for 10 degree
observer.

Results of mapping obtained with paper B (HP Professional Matte Canvas) under different
illuminats.

Figure 63: Color samples mapped to gamut of HP Professional Matte Canvas under illuminant D50 for 10
degree observer.

Results of mapping obtained with paper C (HP Artist Matte Canvas) under different illumi-
nats.

47
 Spectral Print Reproduction - Modeling and Feasibility

Figure 64: Color samples mapped to gamut of HP Professional Matte Canvas under illuminant D65 for 10
degree observer.

Figure 65: Color samples mapped to gamut of HP Professional Matte Canvas under illuminant A for 10
degree observer.

Figure 66: Color samples mapped to gamut of HP Professional Matte Canvas under illuminant C for 10
degree observer.

Figure 67: Color samples mapped to gamut of HP Professional Matte Canvas under illuminant F2 for 10
degree observer.

Figure 68: Color samples mapped to gamut of HP Artist Matte Canvas under illuminant D50 for 10 degree
observer.

Figure 69: Color samples mapped to gamut of HP Artist Matte Canvas under illuminant D65 for 10 degree
observer.

Figure 70: Color samples mapped to gamut of HP Artist Matte Canvas under illuminant A for 10 degree
observer.

Figure 71: Color samples mapped to gamut of HP Artist Matte Canvas under illuminant C for 10 degree
observer.

Figure 72: Color samples mapped to gamut of HP Artist Matte Canvas under illuminant F2 for 10 degree
observer.

Results of mapping obtained with paper D (HP Artist Matte Canvas) under different illumi-
nats.

Results of mapping obtained with paper E (HP Recycled Bond Paper) under different illumi-

48
 Spectral Print Reproduction - Modeling and Feasibility

Figure 73: Color samples mapped to gamut of HP Artist Matte Canvas under illuminant D50 for 10 degree
observer.

Figure 74: Color samples mapped to gamut of HP Universal Coated Paper under illuminant D65 for 10
degree observer.

Figure 75: Color samples mapped to gamut of HP Universal Coated Paper under illuminant A for 10 degree
observer.

Figure 76: Color samples mapped to gamut of HP Universal Coated Paper under illuminant C for 10 degree
observer.

Figure 77: Color samples mapped to gamut of HP Universal Coated Paper under illuminant F2 for 10 degree
observer.

nats.

Figure 78: Color samples mapped to gamut of HP Recycled Bond Paper under illuminant D50 for 10 degree
observer.

Figure 79: Color samples mapped to gamut of HP Recycled Bond Paper under illuminant D65 for 10 degree
observer.

Figure 80: Color samples mapped to gamut of HP Recycled Bond Paper under illuminant A for 10 degree
observer.

Figure 81: Color samples mapped to gamut of HP Recycled Bond Paper under illuminant C for 10 degree
observer.

49
 Spectral Print Reproduction - Modeling and Feasibility

Figure 82: Color samples mapped to gamut of HP Recycled Bond Paper under illuminant F2 for 10 degree
observer.

50
 Spectral Print Reproduction - Modeling and Feasibility

8 APPENDIX B

Ini!al Mapped
Na me L a b L a b
BLUE I 39,39 -1,15 -49,35 39,21569 0 -48
BLACK 12,91 0,09 -2,09 17,64706 0 0
BLUE II 21,01 -0,22 -28,03 27,05882 0 -25
CHERRY 28,05 19,97 9,29 28,23529 20 9
YELLOW 84,57 -5,08 80,11 84,70588 -4 80
BLUE III 25,91 3,1 -43,59 31,76471 2 -41
CORAL 68,38 43,07 40,52 60,39216 39 37
CORAL II 79,49 19,51 7,72 79,21569 19 7
CREMA 95,47 1,98 7,34 90,58824 1 6
YELLOW II 89,27 10,82 89,64 76,47059 9 80
DUNE 68,81 2,98 14,59 68,62745 3 15
BROWN 24,87 4,13 8,96 24,70588 4 9
RED 51,86 54,02 51,33 48,62745 49 46
BLUE IV 79,3 -19,86 -13,76 78,82353 -18 -12
YELLOW III 78 12,32 51,77 78,03922 12 52
GREEN 45,55 -34,29 14,42 45,4902 -33 14
YELLOW IV 62,69 10,51 54,2 62,7451 11 54
PURPLE 33,79 33,39 -37,37 33,72549 33 -36
MAGENTA 36,82 40,09 -5,46 36,86275 40 -4
MAGENTA II 28,67 25,6 -7,4 28,62745 26 -6
RED I 61,51 52,68 57,58 52,54902 44 48
BLUE V 15,71 0,19 -7,05 19,60784 0 -4
ORANGE 57,06 49,97 64,23 54,11765 41 53
PURPLE I 17,84 6,88 -35,55 27,45098 6 -32
YELLOW V 72,49 15,56 79,09 71,76471 15 76
PINK 75,65 8,51 12,21 75,68627 9 12
GREEN I 86,83 -14,48 53,93 86,66667 -13 54
BROWN I 74,02 6,3 12,7 74,11765 6 13
BLACK I 47,27 4,83 10,09 47,45098 5 10
RED II 84,75 13,9 41,72 79,60784 13 40
GREEN II 35,83 -23,17 8,81 35,68627 -22 9
YELLOW VI 78,19 5,52 83,84 78,03922 6 84

Figure 83: CIELAB values of the target and mapped to gamut created with paper A for UL300 illuminant
and 10 degrees observer

51
 Spectral Print Reproduction - Modeling and Feasibility

Ini!al Mapped
Na me L a b L a b
BLUE I 39,39 -1,15 -49,35 39,21569 0 -48
BLACK 12,91 0,09 -2,09 19,21569 0 0
BLUE II 21,01 -0,22 -28,03 27,84314 0 -24
CHERRY 28,05 19,97 9,29 28,23529 20 9
YELLOW 84,57 -5,08 80,11 84,70588 -4 80
BLUE III 25,91 3,1 -43,59 31,76471 2 -41
CORAL 68,38 43,07 40,52 68,23529 43 41
CORAL II 79,49 19,51 7,72 79,60784 20 8
CREMA 95,47 1,98 7,34 93,33333 2 7
YELLOW II 89,27 10,82 89,64 82,7451 10 88
DUNE 68,81 2,98 14,59 68,62745 3 15
BROWN 24,87 4,13 8,96 24,70588 4 9
RED 51,86 54,02 51,33 51,76471 54 51
BLUE IV 79,3 -19,86 -13,76 79,21569 -19 -13
YELLOW III 78 12,32 51,77 78,03922 12 52
GREEN 45,55 -34,29 14,42 45,4902 -33 14
YELLOW IV 62,69 10,51 54,2 62,7451 11 54
PURPLE 33,79 33,39 -37,37 33,72549 33 -36
MAGENTA 36,82 40,09 -5,46 36,86275 40 -4
MAGENTA II 28,67 25,6 -7,4 28,62745 26 -6
RED I 61,51 52,68 57,58 61,56863 53 58
BLUE V 15,71 0,19 -7,05 20,78431 0 -4
ORANGE 57,06 49,97 64,23 57,2549 50 64
PURPLE I 17,84 6,88 -35,55 27,45098 6 -32
YELLOW V 72,49 15,56 79,09 72,54902 16 79
PINK 75,65 8,51 12,21 75,68627 9 12
GREEN I 86,83 -14,48 53,93 86,66667 -13 54
BROWN I 74,02 6,3 12,7 74,11765 6 13
BLACK I 47,27 4,83 10,09 47,45098 5 10
RED II 84,75 13,9 41,72 84,70588 14 42
GREEN II 35,83 -23,17 8,81 35,68627 -22 9
YELLOW VI 78,19 5,52 83,84 78,03922 6 84

Figure 84: CIELAB values of the target and mapped to gamut created with paper B for UL300 illuminant
and 10 degrees observer

52
 Spectral Print Reproduction - Modeling and Feasibility

Ini!al Mapped
Na me L a b L a b
BLUE I 39,39 -1,15 -49,35 39,21569 0 -48
BLACK 12,91 0,09 -2,09 18,03922 0 0
BLUE II 21,01 -0,22 -28,03 27,45098 0 -24
CHERRY 28,05 19,97 9,29 28,23529 20 9
YELLOW 84,57 -5,08 80,11 84,70588 -4 80
BLUE III 25,91 3,1 -43,59 31,76471 2 -41
CORAL 68,38 43,07 40,52 68,23529 43 41
CORAL II 79,49 19,51 7,72 79,60784 20 8
CREMA 95,47 1,98 7,34 94,5098 1 6
YELLOW II 89,27 10,82 89,64 83,92157 10 89
DUNE 68,81 2,98 14,59 68,62745 3 15
BROWN 24,87 4,13 8,96 24,70588 4 9
RED 51,86 54,02 51,33 51,76471 54 51
BLUE IV 79,3 -19,86 -13,76 79,21569 -19 -13
YELLOW III 78 12,32 51,77 78,03922 12 52
GREEN 45,55 -34,29 14,42 45,4902 -33 14
YELLOW IV 62,69 10,51 54,2 62,7451 11 54
PURPLE 33,79 33,39 -37,37 33,72549 33 -36
MAGENTA 36,82 40,09 -5,46 36,86275 40 -4
MAGENTA II 28,67 25,6 -7,4 28,62745 26 -6
RED I 61,51 52,68 57,58 61,56863 53 58
BLUE V 15,71 0,19 -7,05 20 0 -4
ORANGE 57,06 49,97 64,23 57,2549 50 64
PURPLE I 17,84 6,88 -35,55 27,45098 6 -32
YELLOW V 72,49 15,56 79,09 72,54902 16 79
PINK 75,65 8,51 12,21 75,68627 9 12
GREEN I 86,83 -14,48 53,93 86,66667 -13 54
BROWN I 74,02 6,3 12,7 74,11765 6 13
BLACK I 47,27 4,83 10,09 47,45098 5 10
RED II 84,75 13,9 41,72 84,70588 14 42
GREEN II 35,83 -23,17 8,81 35,68627 -22 9
YELLOW VI 78,19 5,52 83,84 78,03922 6 84

Figure 85: CIELAB values of the target and mapped to gamut created with paper C for UL300 illuminant
and 10 degrees observer

53
 Spectral Print Reproduction - Modeling and Feasibility

Ini!al Mapped
Na me L a b L a b
BLUE I 39,39 -1,15 -49,35 39,21569 0 -48
BLACK 12,91 0,09 -2,09 15,68627 0 0
BLUE II 21,01 -0,22 -28,03 27,05882 0 -24
CHERRY 28,05 19,97 9,29 28,23529 20 9
YELLOW 84,57 -5,08 80,11 84,70588 -4 80
BLUE III 25,91 3,1 -43,59 33,33333 2 -39
CORAL 68,38 43,07 40,52 68,23529 43 41
CORAL II 79,49 19,51 7,72 79,60784 20 8
CREMA 95,47 1,98 7,34 92,15686 1 6
YELLOW II 89,27 10,82 89,64 80 10 87
DUNE 68,81 2,98 14,59 68,62745 3 15
BROWN 24,87 4,13 8,96 24,70588 4 9
RED 51,86 54,02 51,33 51,76471 54 51
BLUE IV 79,3 -19,86 -13,76 79,21569 -19 -13
YELLOW III 78 12,32 51,77 78,03922 12 52
GREEN 45,55 -34,29 14,42 45,4902 -33 14
YELLOW IV 62,69 10,51 54,2 62,7451 11 54
PURPLE 33,79 33,39 -37,37 33,72549 33 -36
MAGENTA 36,82 40,09 -5,46 36,86275 40 -4
MAGENTA II 28,67 25,6 -7,4 28,62745 26 -6
RED I 61,51 52,68 57,58 60,78431 52 57
BLUE V 15,71 0,19 -7,05 18,03922 0 -4
ORANGE 57,06 49,97 64,23 59,21569 49 63
PURPLE I 17,84 6,88 -35,55 28,23529 6 -30
YELLOW V 72,49 15,56 79,09 72,54902 16 79
PINK 75,65 8,51 12,21 75,68627 9 12
GREEN I 86,83 -14,48 53,93 86,66667 -13 54
BROWN I 74,02 6,3 12,7 74,11765 6 13
BLACK I 47,27 4,83 10,09 47,45098 5 10
RED II 84,75 13,9 41,72 83,92157 13 41
GREEN II 35,83 -23,17 8,81 35,68627 -22 9
YELLOW VI 78,19 5,52 83,84 78,03922 6 84

Figure 86: CIELAB values of the target and mapped to gamut created with paper D for UL300 illuminant
and 10 degrees observer

54
 Spectral Print Reproduction - Modeling and Feasibility

Ini!al Mapped
Na me L a b L a b
BLUE I 39,39 -1,15 -49,35 47,45098 0 -43
BLACK 12,91 0,09 -2,09 21,17647 0 2
BLUE II 21,01 -0,22 -28,03 33,72549 0 -19
CHERRY 28,05 19,97 9,29 31,37255 18 8
YELLOW 84,57 -5,08 80,11 83,92157 -4 80
BLUE III 25,91 3,1 -43,59 41,56863 2 -34
CORAL 68,38 43,07 40,52 63,52941 41 39
CORAL II 79,49 19,51 7,72 79,21569 19 7
CREMA 95,47 1,98 7,34 87,84314 1 6
YELLOW II 89,27 10,82 89,64 76,86275 8 71
DUNE 68,81 2,98 14,59 68,62745 3 15
BROWN 24,87 4,13 8,96 26,27451 3 7
RED 51,86 54,02 51,33 58,82353 45 42
BLUE IV 79,3 -19,86 -13,76 78,43137 -18 -12
YELLOW III 78 12,32 51,77 78,03922 12 52
GREEN 45,55 -34,29 14,42 45,4902 -33 14
YELLOW IV 62,69 10,51 54,2 63,52941 10 53
PURPLE 33,79 33,39 -37,37 41,56863 30 -33
MAGENTA 36,82 40,09 -5,46 40,39216 38 -3
MAGENTA II 28,67 25,6 -7,4 33,33333 23 -5
RED I 61,51 52,68 57,58 60,78431 41 45
BLUE V 15,71 0,19 -7,05 23,13725 0 -1
ORANGE 57,06 49,97 64,23 62,35294 37 48
PURPLE I 17,84 6,88 -35,55 36,07843 5 -25
YELLOW V 72,49 15,56 79,09 74,5098 13 67
PINK 75,65 8,51 12,21 75,68627 9 12
GREEN I 86,83 -14,48 53,93 86,66667 -13 54
BROWN I 74,02 6,3 12,7 74,11765 6 13
BLACK I 47,27 4,83 10,09 47,45098 5 10
RED II 84,75 13,9 41,72 79,21569 13 40
GREEN II 35,83 -23,17 8,81 35,68627 -21 8
YELLOW VI 78,19 5,52 83,84 78,82353 5 74

Figure 87: CIELAB values of the target and mapped to gamut created with paper E for UL300 illuminant
and 10 degrees observer

55
 Spectral Print Reproduction - Modeling and Feasibility

9 APPENDIX C

Hints in calculation process using Matlab.

%I n v e r s e Murray−D a v i e s f o r e s t i m a t i n g e f f e c t i v e c o v e r a g e o f each c o l o r a n t

%%%%%%%%%%%%%%%
% C+M+Y+K %
%%%%%%%%%%%%%%%

o 4 _ r e f l=o3_ramp ;
r _ p a p e r=( o 4 _ r e f l ( 5 7 , 5 : 4 0 ) ) ;
c100=o 4 _ r e f l ( 1 , 5 : 4 0 ) ;
m100=o 4 _ r e f l (15 , 5 : 4 0 ) ;
y100=o 4 _ r e f l (29 , 5 : 4 0 ) ;
k100=o 4 _ r e f l (43 , 5 : 4 0 ) ;

% CYAN E f f e c t i v e a r e a
f o r i =1:14
c_measured ( i , : ) = ( o 4 _ r e f l ( i , 5 : 4 0 ) − o 4 _ r e f l ( 5 7 , 5 : 4 0 ) ) ;
end

R_t=( o 4 _ r e f l ( 1 , 5 :4 0) − r _ p a p e r ) ;

f o r i =1:14
upper_c ( i , : ) = ( c_measured ( i , : ) ∗ R_t ( 1 , : ) ’ ) ;
end

R _ t t=R_t ’ ;
lower_c= R_t ∗ R _ t t ;

C _ e f f=upper_c . / lower_c ;
C _ e f f=C _ e f f ( 1 : 1 4 , : ) ;

% CMYK_eff = [ C _ e f f ( 1 : 1 4 , : ) ] ;

c _ t h e o =o 4 _ r e f l ( 1 : 1 4 , 1 ) / 1 0 0 ;
f i g u r e (1)
p l o t ( c_theo , C _ e f f , ’ c ’ ) ;

% MAGENTA E f f e c t i v e a r e a
o4_refl_m=o 4 _ r e f l ( 1 5 : 2 8 , : ) ;
f o r i =1:14

56
 Spectral Print Reproduction - Modeling and Feasibility

m_measured ( i , : ) = ( o4_refl_m ( i , 5 : 4 0 ) − o 4 _ r e f l ( 5 7 , 5 : 4 0 ) ) ;
end

M_R_t=(o4_refl_m ( 1 , 5:4 0) − r _ p a p e r ) ;

f o r i =1:14
upper_m ( i , : ) = ( m_measured ( i , : ) ∗ M_R_t ( 1 , : ) ’ ) ;
end

M_R_tt=M_R_t ’ ;
lower_m= M_R_t∗ M_R_tt ;

M_eff=upper_m . / lower_m ;
M_eff=M_eff ( 1 : 1 4 , : ) ;

% CMYK_eff = [ C _ e f f ( 1 : 3 0 , : ) M_eff ( 1 : 3 0 , 1 ) ] ;

m_theo =o 4 _ r e f l ( 1 : 1 4 , 1 ) / 1 0 0 ;
f i g u r e (2)
p l o t ( m_theo , M_eff ( 1 : 1 4 , 1 ) , ’m’ ) ;

% Yellow E f f e c t i v e a r e a
o 4 _ r e f l _ y=o 4 _ r e f l ( 2 9 : 4 2 , : ) ;
f o r i =1:14
y_measured ( i , : ) = ( o 4 _ r e f l _ y ( i , 5 : 4 0 ) − o 4 _ r e f l ( 5 7 , 5 : 4 0 ) ) ;
end

y_measured = y_measured ( 1 : 1 4 , : ) ;
Y_R_t=( o 4 _ r e f l _ y ( 1 , 5: 40) − r _ p a p e r ) ;

f o r i =1:14
upper_y ( i , : ) = ( y_measured ( i , : ) ∗ Y_R_t ( 1 , : ) ’ ) ;
end

Y _ R _ t t=Y_R_t ’ ;
lower_y= Y_R_t ∗ Y _ R _ t t ;

Y _ e f f=upper_y . / lower_y ;
Y _ e f f=Y _ e f f ( 1 : 1 4 , : ) ;

% CMYK_eff = [ C _ e f f ( 1 : 3 0 , : ) M_eff ( 1 : 3 0 , 1 ) Y_eff (1:30 ,1)] ;

y_theo =o 4 _ r e f l ( 1 : 1 4 , 1 ) / 1 0 0 ;
f i g u r e (3)

57
 Spectral Print Reproduction - Modeling and Feasibility

p l o t ( y_theo , Y _ e f f ( 1 : 1 4 , 1 ) , ’ r ’ ) ;

% BLACK E f f e c t i v e a r e a

o 4 _ r e f l _ k=o 4 _ r e f l ( 4 3 : 5 6 , : ) ;
f o r i =1:14
k_measured ( i , : ) = ( o 4 _ r e f l _ k ( i , 5 : 4 0 ) − o 4 _ r e f l ( 5 7 , 5 : 4 0 ) ) ;
end

k_measured = k_measured ( 1 : 1 4 , : ) ;
K_R_t=( o 4 _ r e f l _ k ( 1 , 5 :40 ) − r _ p a p e r ) ;

f o r i =1:14
upper_k ( i , : ) = ( k_measured ( i , : ) ∗ K_R_t ( 1 , : ) ’ ) ;
end

K_R_tt=K_R_t ’ ;
lower_k= K_R_t ∗ K_R_tt ;

K _ e f f=upper_k . / lower_k ;
K _ e f f=K _ e f f ( 1 : 1 4 , : ) ;
________________________________________________________

%t e s t i n g Neugebauer model with t h e KM measured and e s t i m a t e d Oce & Xerox NPs

%l o a d i n g i l l u m i n a n t s and o b s e r v e r data
%l o a d i n g Neugebauer p r i m e r i e s o f our two p r i n t e r s and
%c o l o r a n t c o n c e n t r a t i o n and r e f l e c t a n c e s o f t h e t e s t c h a r t

NP = l o a d ( ’ x3_eNP_KM . t x t ’ ) ; % f o r Xerox e s t i m a t e d NPs

NP = l o a d ( ’ x3_mNP . t x t ’ ) ; % f o r Xerox measured NPs
NP = l o a d ( ’ O4_eNPs . t x t ’ ) ; % f o r Oce e s t i m a t e d NPs
NP = l o a d ( ’ O4_eNPs . t x t ’ ) ; % f o r Oce measured NPs

Mref = l o a d ( ’ X3_tc . t x t ’ ) ;
Mref = x 3 _ t c ( 1 9 3 : 3 0 0 , 5 : 4 0 ) ;
C = x3_tc (193:300 ,1:4);
%Cc i s i n t e r p o l a t e d with E f f _ C o v e r a g e r e s u l t s

NP=NP ’ ;

f o r i =1:108
ww ( : , i )= demichel2 (C( i , : ) ’ ) ;
end
NP=NP ’ ;

58
 Spectral Print Reproduction - Modeling and Feasibility

f o r i =1:108
r e f ( : , i ) = NP∗ww( : , i ) ;
end

NP = NP ’ ;
f o r i =1:108
ww = demichel2 (C( i , : ) ’ ) ;
r e f ( : , i ) = NP∗ww;
end
Mref = Mref ’ ;
RMSE = calcRMSE ( r e f , Mref ) ;
GFC = calcGFC ( r e f , Mref ) ;
MRMSE = sum(RMSE) / 1 0 8 ;
MGFC = sum(GFC) / 1 0 8 ;

____________________________________________________________

%ESTIMATION o f X3 ramps by MD u s i n g E f f e c t i v e C o l o r Coverage

f o r i =1:14
r _ e s t _ c _ e f f ( i ,:)=(1 −( c ( i , 1 ) / 1 0 0 ) ) ∗ r _ p a p e r+(c ( i , 1 ) / 1 0 0 ) ∗ r_maxcove ;
end

% ESTIMATED Magenta ramps

f o r i =1:14
r _ e s t _ m _ e f f ( i ,:)=(1 −( c ( i , 2 ) / 1 0 0 ) ) ∗ r _ p a p e r+(c ( i , 2 ) / 1 0 0 ) ∗ r_maxcove ;
end

% ESTIMATED Yellow ramps

f o r i =1:14
r _ e s t _ y _ e f f ( i ,:)=(1 −( c ( i , 3 ) / 1 0 0 ) ) ∗ r _ p a p e r+(c ( i , 3 ) / 1 0 0 ) ∗ r_maxcove ;
end

% ESTIATED B l a c k ramps
f o r i =1:14
r _ e s t _ k _ e f f ( i ,:)=(1 −( c ( i , 4 ) / 1 0 0 ) ) ∗ r _ p a p e r+(c ( i , 4 ) / 1 0 0 ) ∗ r_maxcove ;
end

_____________________________________________________________

%t e s t i n g YN−Neugebauer model with t h e KM e s t i m a t e d NPs

f o r paper = 1:5

NP = l o a d ( [ paper { p a p e r i d x } , ’ MNPs . t x t ’ ] ) ;
Mref = l o a d ( [ paper { p a p e r i d x } , ’ TC . t x t ’ ] ) ;
Ramp = l o a d ( [ paper { p a p e r i d x } , ’ S c a l e . t x t ’ ] ) ;
NP = NP ’ ;

r e f = zeros (36 ,1504 ,51);

59
 Spectral Print Reproduction - Modeling and Feasibility

f o r n =1:0.1:maxn ;
n;
a e f f = c o l c o v e r a g e (Ramp.^(1/n ) ) ;
f o r i =1:908
ww = demichel2 ( a e f f ( i , : ) ’ ) ;
r e f ( : , i , u i n t 8 ( ( n −0.9)∗10)) = (NP.^(1/n )∗ww).^n ;
end
end

Mref = Mref ’ ;

n=l e n g t h ( 1 : 0 . 1 : maxn ) ;

f o r j= 1 : n
temp = r e f ( : , : , j ) ;

RMSE = calcRMSE ( temp ( 3 : 3 3 , : ) , Mref ( 3 : 3 3 , : ) ) ;

GFC = calcGFC ( temp ( 3 : 3 3 , : ) , Mref ( 3 : 3 3 , : ) ) ;

RMSEw{ j } = RMSE;
GFCw{ j } = GFC ;
end
f o r i =1:n
ARMSE ( : , i ) = RMSEw{ i } ’ ;
AGFC ( : , i ) = GFCw{ i } ’ ;
end
MRMSE = mean(ARMSE ) ;
stdRMSE = s t d (ARMSE ) ;
maxRMSE = max(ARMSE ) ;

MGFC = mean(AGFC ) ;
stdGFC = s t d (AGFC ) ;
minGFC = min (AGFC ) ;

___________________________________________________________________
j c p _ l a b =read ( j c p _ l a b . t x t ) ;
j c p _ l a b 3 d=r e s h a p e ( j c p _ l a b , 3 2 , 1 , [ ] ) ;
i m w r i t e ( j c p _ l a b 3 d , ’ j c p _ l a b 3 d . t i f f ’ , ’ t i f f ’ , ’ C o l o r s p a c e ’ , ’ c i e l a b ’ , ’ Compression ’ , ’ non
mapped=imread ( ’ mapped_X ’ , ’ t i f f ’ ) ;

% t = T i f f ( ’ mapped_A ’ , ’ r ’ ) ;
% [ L a b ] = t . read ( ) ;
% L = double ( L ) ∗ ( 1 0 0 / 2 5 5 ) ;
% a = double ( a ) − 128;
% b = double ( b ) − 128;
%i m f i n f o ( ’ mappedT ’ , ’ t i f f ’ ) ;
%mapped_lab2d=r e s h a p e ( mapped_lab3d , 3 3 , 3 ) ;
% f i n a l= i c c l a b 2 c i e l a b ( ’ mapped_lab2d ’ ) ;

60
 Spectral Print Reproduction - Modeling and Feasibility

% i n p u t _ c l a s s = c l a s s ( ’ proba . t i f f ’ ) ;
% S = char ( i n p u t _ c l a s s ) ;
% C= double ( i n p u t _ c l a s s ) ;
% D=double (C ) ;

i n p u t _ c l a s s = c l a s s ( mapped ) ;
c i e l a b = double ( mapped ) ;

i f strcmp ( i n p u t _ c l a s s , ’ uint8 ’ )
% F i x L∗ v a l u e s .
cielab (: ,: ,1) = cielab (: ,1)/2.55;
c i e l a b ( : , : , 2 ) = c i e l a b (: ,2) −128;
c i e l a b ( : , : , 3 ) = c i e l a b (: ,3) −128;
end

mapped2d=r e s h a p e ( c i e l a b , 3 2 , 3 ) ;
mapped2d=double ( mapped2d ) ;

f o r i =1:32
DE( i )=d E c a l c ( j c p _ l a b ( i , 1 : 3 ) , mapped2d ( i , 1 : 3 ) ) ;
DE94( i ) = CIE94 ( j c p _ l a b ( i , 1 : 3 ) , mapped2d ( i , 1 : 3 ) , 1 , 1 , 1 ) ;
DCMC( i )=cmc ( j c p _ l a b ( i , 1 : 3 ) , mapped2d ( i , 1 : 3 ) ) ;
DE00( i )=deltaE2000 ( j c p _ l a b ( i , 1 : 3 ) , mapped2d ( i , 1 : 3 ) ) ;
end

mapped_lab3d16=u i n t 1 6 ( mapped_lab3d ) ;
i n p u t _ c l a s s = c l a s s ( mapped_lab3d16 ) ;
c i e l a b = double ( mapped_lab3d16 ) ;

i f strcmp ( i n p u t _ c l a s s , ’ uint16 ’ )
% F i x L∗ v a l u e s .
c i e l a b ( : , : , 1 ) = c i e l a b ( : , : , 1 ) ∗ (65535/65280);

% How much t o s h i f t t h e a∗ and b∗ v a l u e s .

s h i f t _ v a l u e = 32768;
else
s h i f t _ v a l u e = 128;
end

i f size ( cielab ,3) > 1

% F i x a∗ and b∗ v a l u e s .
ab = c i e l a b ( : , : , [ 2 3 ] ) ;

% S h i f t low v a l u e s up .
mask = ab <= ( s h i f t _ v a l u e − 1 ) ;
ab ( mask ) = ab ( mask ) + s h i f t _ v a l u e ;

61
 Spectral Print Reproduction - Modeling and Feasibility

% S h i f t high v a l u e s down .
mask = ~mask ;
ab ( mask ) = ab ( mask ) − s h i f t _ v a l u e ;

c i e l a b ( : , : , [ 2 3 ] ) = ab ;
end

i f strcmp ( i n p u t _ c l a s s , ’ uint16 ’ )
c i e l a b = uint16 ( c i e l a b ) ;
else
c i e l a b = uint8 ( c i e l a b ) ;
end

____________________________________________________

62
 Spectral Print Reproduction - Modeling and Feasibility

Bibliography

[1] Assefa, M. Kubelka munk theory for efficient spectral printer modelling. Master’s thesis,
2010.

[2] Blahov. Evaluating the use of mixed neugebauer primaries, measured and estimated by
kubelka-munk theory, in spectral reproduction using spectral neugebauer model and yule-
nielsen modification. Master’s thesis.

[3] Balasubramanian, R. The use of spectral regression in modeling halftone color printers.

[4] Bugnon, T. & Hersch, R. D. Constrained acquisition of ink spreading curves from printed
color images.

[5] Di-Yuan Tzeng, R. S. B. A review of principal component analysis and its applications to
color technology.

[6] Je-Ho Lee, Member, I. & Jan P. Allebach, Fellow, I. Inkjet printer model-based halftoning.

[7] A guide to halftone technologies.

[8] Kipphan, H. 2001. Handbook of Print Media: Technologies and Production Methods.

[9] Murat Mese, P. P. V. Recent advances in digital halftoning and inverse halftoning methods.

[10] Balasubramanian, R. The use of spectral regression in modeling halftone color printers.

[11] Evans, B. L. Error diffusion halftoning methods forimage display.

[12] Derhak, M. Spectral print inversion strategies.

[13] M. J. Vrhel, H. J. T. Color printer characterization in matlab.

[14] Emmel, P., Amidror, I., Ostromoukhov, V., & Hersch, R. D. 2005. Predicting the spectral
behaviour of colour printers for transparent inks on transparent support.

[15] Jeremie Gerhardt, J. Y. H. Spectral color reproduction minimizing spectral and perceptual
color differences.

[16] Gerhardt, J. Reproduction spectrale de la couleur: approches par modelisation dimprimante

et par halftoning avec diffusion derreur vectorielle. PhD thesis.

[17] Fabrice Rousselle, Thomas Bugnon, R. D. H. Spectral prediction model for variable dot-size
printers.

63
 Spectral Print Reproduction - Modeling and Feasibility

[18] Hong-Kee Kim, Byoung-Ho Kang, G.-S. H. J.-S. K. A method of printer modeling with the
spectral reflectance curves using error back propagation.

[19] Zuffi, S. & Schettini, R. 2004. Modeling dot gain and inks interaction. In in Proc. IS&T/SID
12th Color Imaging Conference (IS&T, 181–186.

[20] Zuffi, S., Schettini, R., & Mauri, G. 2005. Spectral-based printer modeling and characteri-
zation. J. Electronic Imaging, 14(2), 023008.

[21] Lawler. 1997. Lawler.

[22] Lewandowski, A., Ludl, M., Byrne, G., & Dorffner, G. Aug 2006. Applying the yule-nielsen
equation with negative n. J. Opt. Soc. Am. A, 23(8), 1827–1834.

[23] Bastani, B. 2009. Spectral analysis of output devices from printing to predicting, PhD thesis.

[24] Lawrenece A. Taplin, R. S. b. Spectral color reproduction based on a six-colorant inkjet

output system.

[25] Binyu Wang, Haisong Xu, M. R. L. & Guo, J. Spectral-based color separation method for a
multi-ink printer.

[26] Urban, P. & Grigat, R.-R. 2006. Spectral-based color separation using linear regression
iteration. Color Research and Application, 31(3), 229–238.

[27] Philipp Urban, M. R. R. R. S. B. Accelerating spectral-based color separation within the

neugebauer subspace.

[28] Balasubramanian, R. A spectral neugebauer model for dot-on-dot printers.

[29] Roy S. Berns, Animesh Bose, D.-Y. T. The spectral modeling of large-format ink-jet printers.

[30] Raimondo Schettini, Daniela Bianucci, G. M.-S. Z. An empirical approach for spectral color
printers characterization.

[31] Roger David Hersch, F. C. Improving the yule-nielsen modified spectral neugebauer model
by dot surface coverages depending on the ink superposition conditions.

[32] Younda Chen, R. S. B. & Taplin, L. A. Six color printer characterization using an optimized
cellular yule-nielsen spectral neugebauer model.

[33] Bastani, B., Cressman, B., Shaw, M., packard Company, H., & Poster, U. 1996. Sparse cellu-
lar neugebauer model for n-ink printers. In Proc. IS&T/SID Fourth Color Imaging Conference:
Color Science, Systems and Applications, 58–60.

[34] D.R.Wyble, R. B. 2000. A critical review of spectral models applied to binary color printing.

[35] Rotea, M. & Lana, C. A robust estimation algorithm for printer modeling.

[36] Yongda Chen, Roy S. Berns, L. A. T. Extending printing color gamut by optimizing the
spectral reflectance of inks.

64
 Spectral Print Reproduction - Modeling and Feasibility

[37] Akao, Y., Yamamoto, A., & Higashikawa, Y. 2009. Estimation of inkjet printer spur gear
teeth number from pitch data string of limited length. In Proceedings of the 3rd Interna-
tional Workshop on Computational Forensics, IWCF ’09, 25–32, Berlin, Heidelberg. Springer-
Verlag.

[38] The 9th Color Imaging Conference: Color Science and Engineering: Systems, Technologies,
Applications, November 6, 2001, Scottsdale, Arizona, USA. IS&T - The Society for Imaging
Science and Technology, 2001.

[39] Binyu Wang, Haisong Xu, M. R. L. a. J. G. Maintaining accuracy of cellular yulenielsen

spectral neugebauer models for different ink cartridges using principal component analysis.

[40] Song-hua Hea, Z. L. The linear colorant mixing space based on kubleka-munk turbid media
theory.

[41] Hunter l,a,b Color Scale.

[42] Taplin, L. A. Spectral modeling of a six-color inkjet printer.

[43] Roy S. Berns, Francisco H. Imai, P. D. B. & Tzeng, D.-Y. Multi-spectral-based color repro-
duction research at the munsell color science laboratory.

[44] Mekides Abebe, J. G. & Hardeberg, J. Y. 2011. Kubelka-munk theory for efficient spectral
printer modeling.

[45] Uribe, J. Understanding black point compensation.

[46] Tzeng, D.-Y. SPECTRAL-BASED COLOR SEPARATION ALGORITHM DEVELOPMENT FOR

MULTIPLE-INK COLOR REPRODUCTION. PhD thesis.

[47] Data Hiding in Halftone Images using Error Diffusion Halftoning with Adaptive Thresholding,
2006.

[48] Bastani, B. SPECTRAL ANALYSIS OF OUTPUT DEVICES, FROM PRINTING TO PREDICTING.

PhD thesis.

[49] Evans, B. L. How to make printed and displayed images have high visual quality.

[50] Shohei Tsutsumi, Mitchell R. Rosen, R. S. B. Spectral color management using interim
connection spaces based on spectral decomposition.

[51] Bugnon, T. & Hersch, R. D. 2011. Illumination estimation via thin-plate spline interpola-
tion. 28, 940–94.

[52] Richard L. Alfvin, M. D. F. Observer variability in metameric color matches using color
reproduction media.

[53] Beretta, G. B. The labpqr color space.

65
 Spectral Print Reproduction - Modeling and Feasibility

[54] Nicolas Bonnier, Francis Schmitt, H. B. & Berche, S. Evaluation of spatial gamut mapping
algorithms.

[55] Mitchell R. Rosen, M. W. D. Spectral gamuts and spectral gamut mapping.

[56] Konica Minolta - Light sources and illuminants.

[57] 2008, January. Introduction of cmc colour differneces, scoop.

[58] Farup, I. & Hardeberg, J. Y. Interactive color gamut mapping.

[59] In-Su Jang, Chang-Hwan Son, T.-Y. P. K.-W. K. & Ha, Y.-H. Hi-fi printer characterization
method using color correlation for gamut extension.

[60] Rosen, M. R. & Ohta, N. 2003. Spectral color processing using an interim connection space.
In Color Imaging Conference, 187–192.

[61] Balasubramanian, R. & Dalal, E. A method for quantifying the color gamut of an output
device.

[62] Adobe photoshop tiff technical notes.

66
 Feasibility of spectral print reproduction

Kristina Marijanovic Jon Yngve Hardeberg

The Norwegian Color and The Norwegian Color and
Media Computing Laboratory Media Computing Laboratory
Gjøvik University College, Gjøvik University College,
Gjøvik, Norway Gjøvik, Norway
[email protected] [email protected]

ABSTRACT reproduced in form of spectral print. For that reason the algorithm
that enables the whole process, from spectral reflectance input to
Within this work answering to question of feasibility of spectral
print reproduction of targeted natural object was aimed. From the final decision of sending to printer or not, automated was proposed.
world that surrounds us human visual system is able to sense much
wider range of color than any output system is able to reproduce. 2. GARNAMENT PRODUCTS
Printing machines are part of that system with their own defined
Within this work focus was on feasibility of particular product. The
environment and communication channels and the real challenge
collaboration with the retail store resulted in obtaining real samples
is these to be overcame. Certain data with different kind of sensors
and testing the possibility of spectral print reproduction with printer
describes the objects, in our case that data is defined with the spec-
in our lab. The challenge was in trying to reduce the cost that re-
trophotometers. Before sending any data to printing process in or-
tailer currently incur on one front of the retail color development
der to obtain spectral print the feasibility of that process should be
process. That particularly involves recurring expenditure on 5 cm
confirmed. If neglecting that fact, hit-and-miss tries would yield big
x 5 cm fabric samples that are produced to within 0.8 DE CMC.
waste of time and energy. For that reason the automated process in
This is valid for illuminant UL3000 and 10 degrees observer. De-
verifying the feasibility of color sensation being printed spectrally
signers use these physical representations when they are deciding
of certain object was created.
which shade an item of garment product will be, or which shade
needs to be placed near another in a printed or yarn dyed pattern
Keywords on a garment product. The existing set of roughly 4000 colors is
Spectral print reproduction, Gamut mapping, Spectral print feasi- supporting the design process within the color set that designers
bility automation are able to pick from. Once their colors for a season and brand
are picked, 5cm x 5cm fabric samples are ordered to work with.
The 4000 colors are supported using spectral data. That reference
1. INTRODUCTION helps to keep the color on shade through many production runs, i.e.
The challenge of spectral print feasibility of target object was tried of color chips, of the color standards, and of the garments that are
to be automated. Instead of several different steps in creation of specified to be in close tolerance of these colors. With that given
that process, one unique was created and the new algorithm, which one set of spectral data for all colors, within scope of these the-
includes all the segments together, and that way saves time, was sis feasibility of printing particular colors out on as needed basis
proposed. In this way big amount of input data can be initially was tested. The possibility of printing these color samples versus
placed and from the output it can be seen exact which parts of the purchasing them from a dyeing company would significantly save
gamut for targeted spectral reproduction are missing. The answer in cost per color. That is why the important step was focus on in-
to set of the spectral print feasibility of number of objects, under the vestigation the feasibility of being able to produce these colors ac-
same illuminant, was given. With the change of the paper substrate curately and consistently over the many variables that arise during
the diversity of possible spectral print reproduction was goal to be consecutive runs. The starting issue in getting the print head cre-
checked. The fact that in process of reaching final confirmation ate the wanted output, was to replace the D50 table with a UL3000
or negation of spectral print possibility many different steps are illuminant. The reason for that was the fact that color standards
conducted it was very important to make that process decreased are made using reactive dyes, and the printers are limited to a sig-
in time consumed. In this particular case there were 32 spectral nificantly smaller set of colorants. Therefore, from a reflectance
data color samples given for which feasibility was to be checked. curve matching standpoint it was expected that there will likely be
Nevertheless, that was just the subset of initial amount wanted to be gaps. To overcome that problem the try to match to UL3000 illu-
minant was very important. Another challenge that was related to
gamut is the difference in available chroma in the chromophores.
Given that there are spectral data for all 4000 colors available, the
brightest colors to be produced to produce are known as well but
only the right combination of paper and colorants will create the
maximum chroma. This is due to having the spectral data for the
targets it would be easy to generate the L*,a*,b* C, and h values
for all colors [2] [6]. From this step then it is possible to deter-
mine the maximum chroma needed. This would be a base ground
for possible realization of coverage for given gamut [4] [7].
3. PROBLEM STATEMENT
What was done by now was the reproduction of the same set of Figure 2: Lab values of color samples.
4000 colors using pigments in an injection molded process. Within
that the knowledge that reflectance of colors made while using re-
active dyes helped in match and use reflectance curves of pigments. 4. EXPERIMENT
That was why the whole process wanted to be moved to one step
further - ink and paper combination. The samples used for our The twelve channel HP DesignJet Z3200 printer was employed in
calculation were subset of original set consisted of 4000 samples. experiment.Having available that strong printing machine was cru-
The subset consisted of 32 data samples. Spectral data of a textile cial element because of its much better gamut size comparing to
standard and a plastic standard for the 10 degree observer weighted the four channel printers. The gamut of the HP DesignJet Z3200
illuminant were given. The colors made by injection molding were printer was obtained for five different papers and visualized using
used as color standards which suppliers in the hard goods area need ICC3D [?] Calibration process of the HP DesignJet Z3200 printer
to match. The idea was to have ’pre-translated’ the color difference results with a printed color chart made of a ramps for each channel.
between textile dyes used in Apparel and Soft home products, and These ramps were then measured manually with spectrophotometer
pigments used in Hard home areas. The main goal was to have the because of the specific hexagonal shape of color patches.
L*a*b* values of printed samples to match the L*a*b* values of
the standard or have the color difference in terms of CMC DE in
UL3000, which is store light, for 10 degree observer between the 5. ALGORITHM
two be as low as possible [5]. An acceptable difference limit was We present the state-of-the art algorithm for making this process
defined with value of is 0.8 DE CMC. Those that fall outside of automated. At the beginning of the proposed algorithm the input
that difference up to 1.2 can be given special consideration by us- data required was the spectral reflectance data measurements. It
ing visual determination of color difference. Those above 1.2 can needs to be noted here that it is important take care of the geom-
be identified as out of gamut for the system under investigation. etry o measurements for the main goal. When (if) spectral print
Spectral reflection data was obtained on spectrophotometer Data- reproduction of wanted object was confirmed to be feasible the fi-
color SF600 Plus CT for each associated color. The values of re- nal print reproduction should be measured under same conditions,
flectance were measured by this instrument at 10 nm interval start- geometry and illuminats. Since the next step was mapping the color
ing at 400nm. Datacolor SF600 plus CT has an integrating sphere gamut of a initial values to the gamut given by output mechanism,
for illumination of the sample and the detector for the spectral ana- printer, the values were represented with an image. For the im-
lyzer is placed at 8 degrees from normal to the surface of the aper- age format Tagged Image File Format was chosen. TIFF image
ture plane (sample port). Sample ports are configurable depending format enables data obtained from spectral reflectances measured
on size of sample being presented. LAV (Large Area View) aper- under certain illuminant to be presented separately for each of L*,
ture plate is 30 mm illuminated and 26nm measured. This is impor- a* and b* channels. If this wanted to be replaced with CMYK val-
tant for ability to maximize the surface area to be measured which ues TIFF is again the best choice because in that case it represents
is ideal to measure color without regard to a textured surface.The four channels separately. By use of TIFF format the conversion
instrument is dual-beam spectrophotometer with diffuse illumina- to RGB space was skipped, regarding important use of displays,
tion and high black trap performance. The main metric that defines during the whole process.This helps in making the whole process
a difference measure used here is CMC. faster and simple. After the TIFF image was created in CIELAB
color space next step was to map that image file, which is the rep-
s
2 2 ∗ 2 resentation of the spectral reflectance of the object wanted to be
L2∗ − L1∗ C2∗ −C1∗
  
∗ ∆Hab printed, to the gamut of a printed for a certain paper. In process
∆ECMC = + + (1)
lSL cSC SH of mapping, ICC3D software was employed and constrains set was
minimum delta and constant hue in order to achieve the best per-
ceptual match of final reproduction. Obtained mapped image is
again in TIFF format which enables to extract the L*, a* and b*
channel data. Next, that data was used in step of comparing the dif-
ferences between the initial TIFF image and mapped TIFF image
[1]. Then the calculation of the differences for several color spaces
in case of several paper substrates under different illuminants was
done. Visualization of the points that are out of gamut for a cer-
tain paper was presented in order to be able to notice which part
of gamut should be considered to be extended if possible was the
next step done. Finally, with the constraints to be considered, i.e.
value more than 1.2 for delta CMC overcomes the aim, the spectral
color points were confirmed or not to be feasible to print within the
gamut possibilities for specific combination of colorants/paper of
printer given. The feasibility of the defined input data from retailer
was then tested with this proposed method for five different papers.
The scheme can be seen in fig. 3.

Figure 1: Spectral reflectances of 32 color samples. 6. RESULTS

 OBJECT
Illuminant Geometry

Spectral reflectance data

[Spectrophotometer] TIFF image Spectral reflectance data
[Spectrophotometer]

Mapping
[ICC3D] Substrate gamut
Constraints
[min Delta, max hue...]

Ini!al TIFF vs. mapped TIFF

Calculated Difference

Feasible
YES NO

Figure 3: Feasibility of spectral print reproduction algorithm

scheme.

With given constraints of the error allowed within delta CMC value Figure 5: 3D gamut visualization of Xerox Phaser 6250 printer is
0.8 the process of spectral color reproduction was decided to be shown on the right side. On the left side L*a*b* values of ramp
started with checking the feasibility. Two printers that were used printed separately for all four channels is presented.
for spectral print modeling were considered to be employed, Xerox
Phaser 6250 color laser and Oce ColorWave 600 printer. The pro-
cess was started with numerical and visual representation of their
gamuts. This was done by using color ramps of separated printed
colorants. Next, spectral reflection data of ramp was measured with
and finally represented with the L*a*b* values by means ICC3D
software. The functionality of the ICC3D software is the ability
to read color data as an image in some standard format or that are
extracted by the gamut tag of an ICC profile [3]. It also enables
the color data to be displayed in different ways in various 3D color
spaces. This is very useful for visualizing the whole process of
checking the spectral print feasibility. The L*a*b* values of the
four channels printer and their gamuts are shown with fig. 4 for
Xerox Phaser 6250 and fig. 4 for the Oce ColorWave 600 printer.

Figure 6: 3D gamut visualization of Oce ColorWave 600 printer is

shown on the right side. On the left side L*a*b* values of ramp
printed separately for all four channels is presented.

Figure 4: 3D gamut visualization of Xerox Phaser 6250 printer is

shown on the right side. On the left side L*a*b* values of ramp
printed separately for all four channels is presented.

Five different papers were employed in obtaining results: HP Heavy

Coated paper (130g/m2 ), HP Professional Matte Canvas (430g/m2 ), Figure 7: 3D gamut visualization of Oce ColorWave 600 printer is
HP Artist Matte Canvas (380g/m2 ), HP Universal Coated Paper shown on the right side. On the left side L*a*b* values of ramp
(95g/m2 ) and HP Recycled Bond Paper (80g/m2 ). After applying printed separately for all four channels is presented.
proposed algorithm the best results were obtained for the HP Artist
Matte Canvas and the worst was given with gamut of HP Recycled
Bond Paper used as substrate. 6.1 HP Artist Matte Canvas
The HP Artist Matte Canvas is 380g/m2 paper often for fine art re- Again, starting from the intimal 3 separated channels in TIFF for-
production. This paper was chosen for the texture that resembles mat the mapped version to gamut obtained on HP Artist Matte Can-
even more to the cloth that the HP Professional Matte Canvas.In vas is created and presented visually with L*a*b* values in color
this particular case of paper it can be seen from the fig. 8 that the ramp of target colors with the fig. 11
maximum CMC differnece is obtained under illuminnat UL3000 so
in comparisment with the rest illuminants the calculations obtained
are satisfied. This is in case of minimum but if the average is to Figure 11: Color samples mapped to gamut of HP Artist Matte
be read, the 1.8 and 0.8 threshold values become overpassed. This Canvas under illuminant UL3000 for 10 degree observer.
is again, as mentioned before, misleading number because the sig-
nificant amount of samples is still above 1.2 values and 0.8. This
can be seen from the histogram in fig. 9. From there is easier to
observe the amount of samples that are feasible to be produced in 6.2 HP Recycled Bond Paper
spectral print reproduction, in particular 20 samples are under CMC The HP Recycled Bond Paper is 80g/m2 paper often for low print
0.8 value and 5 under CMC 1.2 value of total 32 samples. In ob- reproduction. This paper was chosen for establishing the relation
between specific requirements with the paper substrate supporting
HP Ar!st Ma"e Canvas the green technologies.This was to check the results with taking
C ΔE Δ94 ΔCMC ΔE00
min max avg min max avg min max avg min max avg into account the future possibility of spectral print reproduction on
D50
D65
0,21
0,21
11,02
10,65
2,45
1,91
0,11
0,11
10,49
10,11
1,96
1,56
0,13
0,13
18,77
18,07
2,73
2,39
0,10
0,13
7,54
7,25
1,57
1,26
reused paper substrate and in that way making it more affordable
A 0,21 11,02 2,47 0,11 10,49 1,97 0,13 17,05 2,73 0,10 7,54 1,58 and desirable for environment.In this particular case of paper it can
C 0,21 11,04 2,47 0,11 10,50 1,97 0,13 17,77 2,73 0,10 7,50 1,58
F2 0,21 11,02 3,31 0,20 10,49 2,50 0,28 18,77 3,19 0,21 7,54 2,02 be seen from the fig. 12 that the maximum CMC differnece is ob-
UL3000 0,21 10,28 1,93 0,11 9,72 1,57 0,13 17,37 2,36 0,10 6,96 1,27
tained under illuminnat UL3000 so in comparison with the rest il-
Figure 8: Results of gamut mapping to paper C given with different luminants the calculations obtained are satisfied. This is in case of
color differences equations under 6 illuminants. minimum but if the average is to be read, the 1.8 and 0.8 thresh-
old values become overpassed. This is again, as mentioned before,
misleading number because the significant amount of samples is
still above 1.2 values and 0.8. This can be seen from the histogram
in fig. 13. From there is easier to observe the amount of samples
20 that are feasible to be produced in spectral print reproduction, in
20 particular 7 samples are under CMC 0.8 value and 3 under CMC
1.2 value of total 32 samples. From these results it can be con-
15
cluded that quality of the paper strongly affects the feasibility of
10 7 the spectral print reproduction. In observing the values that go
5
5 HP Recycled Bond paper
E ΔE Δ94 ΔCMC ΔE00
min max avg min max avg min max avg min max avg
0 D50 0,26 22,44 7,82 0,21 19,70 5,23 0,28 34,64 6,74 0,25 14,98 4,30
0 - 0.8 0.81 - 1.2 1.2 < D65 0,26 22,56 7,86 0,21 19,32 5,17 0,28 33,95 6,66 0,25 14,67 4,25
A 0,26 24,48 8,53 0,21 19,83 5,58 0,28 34,71 7,13 0,25 15,17 4,62
32 Color Samples C 0,26 22,11 7,64 0,21 19,32 5,06 0,28 33,95 6,55 0,25 14,67 4,16
F2 0,26 30,78 10,09 0,24 18,41 6,15 0,33 32,47 7,61 0,29 14,07 5,10
UL3000 0,26 22,57 7,57 0,21 18,79 5,06 0,28 33,16 6,54 0,25 14,38 4,16

Figure 9: Histogram of resulting CMC difference for 32 color sam-

Figure 12: Results of gamut mapping to paper E given with differ-
ples mapped to printer gamut on paper C.
ent color differences equations under 6 illuminants.

serving the values that go over the 1.2 CMC value, the particular
color points that are the most far away from wanted values to be
mapped are in yellow and dark blue(black) part of the color gamut
22
that can be seen from the fig. 10. From here the lead to direction 25

to look for gamut extension in order to achieve the wanted spectral 20

reproduction is created.
15
7
10
3
5

0
0 - 0.8 0.81 - 1.2 1.2 <

32 Color Samples

Figure 13: Histogram of resulting CMC difference for 32 color

samples mapped to printer gamut on paper E.

over the 1.2 CMC value, the particular color points that are the
most far away from wanted values to be mapped are in yellow and
dark blue(black) part of the color gamut that can be seen from the
Figure 10: Visualization of initial color values and gamut of print fig. 14. From here the lead to direction to look for gamut exten-
created with paper C before mapping, in 3D space. sion in order to achieve the wanted spectral reproduction is created.
Starting from the intimal 3 separated channels in TIFF format the
Initial
A
B
C
D
E

Figure 17: Color samples mapped to gamut of five different papers

under illuminant UL3000 for 10 degree observer. The arrows point
the samples most affected with the gamut mapping. Initial L*a*b*
values are given in top ramp followed by papers A,B,C,D,and E
towards the bottom. The arrows point the samples most affected
with the gamut mapping.

Figure 14: Visualization of initial color values and gamut of print

created with paper E before mapping, in 3D space. 8. CONCLUSION
We presented a more way of checking the possibility weather the
mapped version to gamut obtained on HP Artist Matte Canvas is certain object and its spectral reflectance data can be printed or not
created and presented visually with L*a*b* values in color ramp of satisfying certain allowed deviations of the perfect spectral match.
target colors with the fig. 15 The new algorithm for making the whole process saving time was
proposed. This is due to including all the segments of the process
into one code. For this particular case the results have shown fea-
Figure 15: Color samples mapped to gamut of HP Recycled Bond sibility of spectral print reproduction more more than half of the
Paper under illuminant UL3000 for 10 degree observer. input data given. The lack was noticed in yellow and blue regions
of the gamut. This is area to be taken into consideration while
adding or changing different colorants. The goal was to make the
process of deciding on feasibility automated for bigger amount of
7. ANALYSIS OF THE RESULTS input data in order to help the whole process of spectral printing be
After all substrates were tested and the feasibility was given with more efficient and simple.
the results obtained the comparison of the five papers was done.
This can be seen from the histogram in fig. 16. The fact that calcu- 9. ACKNOWLEDGMENTS
lating the average difference from the ramp of target set of colors
to be spectrally reproduced is not giving the correct answer need We want to thank the Norwegian Color and Media Computing Lab-
to be noted. For that reason histogram representation was used for oratory for providing the resources for the research and retailer
more thorough conclusions about the feasibility. This move brings store for providing us with samples and data. We also thank Eu-
the improvement to visual representation of the data as well as to ropean Commission for funding Color in Informatics and Media
the calculation results. Finally, all of the mapped values are given Technology (CIMET) masters program.

10. REFERENCES
25
[1] Adobe photoshopő tiff technical notes.
20
[2] Hunter l,a,b Color Scale.
[3] I. Farup and J. Y. Hardeberg. Interactive color gamut mapping.
15 0 - 0.8 [4] H. Kipphan. Handbook of Print Media: Technologies and
0.81 - 1.2 Production Methods. 2001.
10 1.2 < [5] H. B. Nicolas Bonnier, Francis Schmitt and S. Berche.
5
Evaluation of spatial gamut mapping algorithms.
[6] P. D. B. Roy S. Berns, Francisco H. Imai and D.-Y. Tzeng.
0 Multi-spectral-based color reproduction research at the
paper A papaer B paper C paper D paper E
munsell color science laboratory.
[7] L. A. T. Yongda Chen, Roy S. Berns. Extending printing color
Figure 16: Histogram of resulting CMC difference for initial 32 gamut by optimizing the spectral reflectance of inks.
color samples mapped to printer gamut on papers A,B,C,D and E.

with the L*a*b* values that are possible to be reproduced. Observ-

ing it carefully gives the opportunity to apply the 3 visualization of
gamuts to real samples wherease the wanted extention of gamut in
yellow and blue (black) direction of gamut space is as well noticed
from fig. ??. If focused on the yellow and blue samples of given
ramp the deviation from the initial ramp can be seen even with the
bare eye. It is somewhat more difficult to noticed yellow mapped
samples difference that blue ones due to ability of human visual
system.