Professional Documents
Culture Documents
Modeling Ink Spreading For Color Prediction: Patrick Emmel
Modeling Ink Spreading For Color Prediction: Patrick Emmel
i
= /6 + (i 1) /3. The sha pe of t he impa ct is a p-
pr oxima t ed by a pa r a met r ic cur ve t ha t joins t he ver t i-
ces defined by t he six vect or s.
Let us denot e = r() t he equa t ion in pola r coor di-
na t es of t his pa r a met r ic cur ve. Beca use we a ssume t ha t
a neighbor influences only loca lly t he sha pe of t he im-
pa ct , t he pa r a met r ic cur ve depends only on r
i
a nd r
i+1
when
i
i+1
. I n or der t o get r ea list ic impa ct sha pes,
we int er pola t e t he va lues of t he r a dius bet ween r
i
a nd
r
i+1
by using a polynomia l of degr ee t hr ee:
= r() = 2(r
i
r
i+1
)t
3
3(r
i
r
i+1
)t
2
+ r
i
(1)
wher e
t i
|
(
'
`
J
J
3
6
a nd
6
1
3 6 3
+ + ( ) . i i
Not e t ha t
r i r
i
6 3
+
|
(
'
`
J
J ,
a nd t ha t t he der iva t ive
dr
d
i
6 3
0 +
|
(
'
`
J
J .
These pr oper t ies gua r a nt ee t he cont inuit y of r() a nd of
it s fir st der iva t ive.
We a ssume t ha t t he densit y D a t t he loca t ion defined
by t he pola r coor dina t es (, ) (0 r()) is given by:
D D
r
M
( , )
( )
|
(
'
`
J
J
]
]
]
]
1
2
(2)
wher e D
M
is t he densit y a t t he cent er of t he impa ct . Not e
t ha t a cir cula r impa ct (r() const a nt ) ha s a pa r a bolic
densit y pr ofile a s obser ved in our pr evious st udy.
4
Th e a mou n t of dye r ema i n s con s t a n t du r i n g t h e
s pr ea di n g pr oces s , on l y t h e s pa t i a l di s t r i bu t i on i s
cha nged. Ther efor e, t he ma xima l densit y D
M
a t t he cen-
t er of t he impa ct must decr ea se when t he impa ct is en-
la r ged. By int egr a t ing Eq. (2) over t he a r ea occupied by
t he impa ct , we get t he t ot a l a mount of dye wit hin t he
impa ct (see Appendix). Beca use t his a mount must r e-
ma in const a nt , we get t he following equa t ion for D
M
:
D D
r
r r r r
M
i
i i i i
+
]
]
]
|
(
'
'
'
'
`
J
J
J
J
J
+ +
0
0
2
1
6
1 1
2
6
13
140
( )
(3)
(a ) (b)
Figure 1. Micr oscopic views of a ha lft oned cya n sa mple (a ) a nd of a gr een ha lft oned sa mple (b). The gr een sa mple (b) is ma de of
t he sa me cya n la yer a s (a ) a nd cover ed wit h a unifor m yellow la yer. Not e t he enla r gement of t he cya n clust er s in (b).
Cyan clust er Paper Gr een clust er Yellow ink
Modeling Ink S preading for Color Prediction Vol. 46, No. 3, May/J une 2002 239
each impact has six neighbor s whose cent er s ar e t he cor -
ner s of a r egular hexagon. In a t hr ee-ink color pr int , each
pixel of t he pr int ed sur face is in one of t he following four
st at es: no ink, cover ed wit h one ink dr op, cover ed wit h
t wo ink dr ops or cover ed wit h t hr ee ink dr ops. Because
an impact has six neighbor s, t her e ar e 4
6
= 4096 possible
neighbor configur at ions. On t he consider ed cent er point ,
a new ink dr op is pr int ed on a sur face t hat is in one of
t he following t hr ee st at es: dr y (no ink), cover ed wit h one
ink dr op or cover ed wit h t wo ink dr ops. Ther efor e, t her e
ar e 3 4
6
= 12288 configur at ions t o consider.
Among t hese 12288 configur a t ions, ma ny a r e equiva -
lent by r eflect ion or by r ot a t ion. Not e t ha t t her e a r e
Figure 2. High r esolut ion gr id modeling t he pr int ed sur fa ce.
The va lue of a gr id point cor r esponds t o t he loca l a mount of
dye. The densit y pr ofile of a n isola t ed ink impa ct is pa r a bolic.
Figure 3. Six vect or s define t he cir cumfer ence of t he impact .
The dashed cir cles indicat e t he locat ions of neighbor ing impact s.
wher e r
0
and D
0
ar e r espect ively t he r adius and t he maxi-
ma l densit y a t t he cent er of a n isola t ed cir cula r impa ct
which did not spr ea d.
Th is n ew impa ct model a llows u s t o s imu la t e t h e
spr ea ding by cha nging t he six r
i
coefficient s a ccor ding
t o t he configur a t ion of t he neighbor s a nd t he st a t e of
t he sur fa ce.
Plya Counting
I n t his sect ion, we der ive t he number of non-equiva -
lent ink-dr op configur a t ions wit h t he help of Plya s
count ing t heor y. (Rea der s int er est ed ma inly in t he fi-
na l ink spr ea ding model ma y skip over t his sect ion.)
Plya s count ing r equir es t hr ee st eps: fir st , defining t he
gr oup of symmet r ies a ct ing on t he set of cor ner s, sec-
ond, fa ct or izing ea ch per mut a t ion int o cycles, a nd t hir d,
ca lcula t ing t he cycle index polynomia l.
As we point ed out in t he pr evious sect ion, most ink jet
pr int er s use a hexagonal gr id when pr int ing in color. So
Figure 4. Ther e a r e six r ot a t ions a nd six r eflect ions t ha t br ing
a r egula r hexa gon ont o it self.
240 J our na l of I ma ging Science a nd Technology
0
6
[a], [b], [c], [d], [e], [f], z
6
1
1
6
[a,b,c,d,e,f] z
6
2
6
[a,c,e], [b,d,f] z
2
3
3
6
[a,d], [b,e], [c,f] z
3
2
4
6
[a,e,c], [b,f,d] z
2
3
5
6
[a,f,e,d,c,b] z
6
ad
[a], [d], [b,f], [c,e] z
2
1
z
2
2
be
[b], [e], [a,c], [d,f] z
2
1
z
2
2
cf
[c], [f], [a,e], [b,d] z
2
1
z
2
2
ab
[a,b], [c,f], [d,e] z
3
2
bc
[b,c], [a,d], [e,f] z
3
2
cd
[c,d], [a,f], [b,e] z
3
2
Modeling Ink S preading for Color Prediction Vol. 46, No. 3, May/J une 2002 241
These set s of ca ses a r e sma ll enough for per for ming
exha ust ive mea sur ement s in or der t o ca libr a t e t he ink
spr ea ding model. Fr om ea ch ca se, we ca n deduce t he
va lue of a coefficient r
i
, which is used by our pr eviously
defined impa ct model. The sha pe of a simula t ed impa ct
depends on six r
i
coefficient s which a r e r ela t ed t o t he
configur a t ion of t he neighbor ing impa ct s a nd t he st a t e
of t he sur fa ce.
Enumerating the Configurations
Plya s count ing t heor y gives t he number of configur a -
t ions t ha t must be consider ed in or der t o ca libr a t e t he
ink spr ea ding model. The configur a t ions t hemselves a r e
comput ed by ot her mea ns. Power ful gener a t ing a lgo-
r it hms exist ,
14
but for t he sake of simplicit y, a naive sieve
met hod is used. A comput er gener a t es t he list of a ll con-
fi gu r a t i on s . Con s i der i n g t h e fi r s t con fi gu r a t i on , a l l
equivalent configur at ions ar e r emoved fr om t he list . The
comput er a pplies t his pr ocedur e t o t he next configur a -
t ion of t he list unt il t he end of t he list is r ea ched. The
30 non-equiva lent configur a t ions for t hr ee-ink-pr int ing
a r e list ed in Fig. 6.
Prediction Results and Discussion
The new ink spr ea ding model wa s combined wit h our
color pr edict ion met hod
2
in or der t o pr edict t he spect r a
of t wo ser ies of 125 sa mples unifor mly dist r ibut ed in
t h e CMY color spa ce. Ea ch ser ies is a set of 5 25
sa mples pr int ed on five sheet s of pa per. The a r ea cover -
a ge of t he yellow ink is const a nt for a ll sa mples pr int ed
on t he sa me sheet , a nd va r ies fr om sheet t o sheet . The
fir st ser ies wa s pr int ed on Epson Glossy Phot o Qua lit y
Pa per using a n Epson St ylus-Color pr int er which is
ba sed on a piezo elect r ic t echnology,
15
a nd t he second
ser ies wa s pr int ed on HP Phot o Pa per, using a n HP-
DJ 560C pr int er which is ba sed on convent iona l t her -
mal ink jet t echniques.
16
All samples wer e pr oduced wit h
a clust er ed dit her a lgor it hm wit h 33 t one levels.
For bot h ser ies, t he r a dius r
0
of a n isola t ed cir cula r
impa ct wa s mea sur ed a ccur a t ely under t he micr oscope.
In t he HP ser ies, t he super posed cya n, ma gent a a nd yel-
low ink dr op impact s have almost t he same size, wher eas
in t he Epson ser ies, t he r a dius of t he yellow dr op im-
pa ct is 20% la r ger t ha n t he r a dius of t he cya n dr op im-
pa ct , which is, in t ur n, is 30% la r ger t ha n t he r a dius of
t he ma gent a dr op impa ct .
Figure 5. A simplified geomet r y: t he hexa gon is subdivided
int o six t r ia ngles t ha t sha r e t he cent er of t he hexa gon a s a
common ver t ex. The spr ea ding of t he ink dr op impa ct in t he
dir ect ion of t he media t or of [a, b] depends only on t he st a t e of
O, a, a nd b.
TABLE II. Cycle Factorizations and Monomials of the Group
S of Symmetries Acting on the Segment [a,b]
Symmetry Cycle factorization Monomial
0
[a], [b]
z
1
2
ab
[a,b]
z
2
Figure 6. List of t he 30 non-equiva lent configur a t ions in a
t hr ee-ink pr int . The da shed cir cles indica t e t he loca t ions of
neighbor s t ha t a r e not cover ed wit h ink. The whit e disks cor -
r espond t o one ink dr op, t he gr a y disks cor r espond t o t wo ink
dr ops, a nd t he bla ck disks cor r espond t o t hr ee ink dr ops. The
vect or s indica t e t he or ient a t ion of t he spr ea ding. The t hir d
a nd sixt h columns indica t e which t est -sa mple is used t o det er -
mine t he spr ea ding coefficient .
242 J our na l of I ma ging Science a nd Technology
E E
94
E
n
94
2
E
94
0% 1.67 1.97 3.58 0.95 1.14 2.49
25% 1.81 1.96 3.11 1.10 1.22 2.37
50% 1.95 2.14 4.13 1.24 1.34 2.46
75% 2.91 3.02 5.59 1.85 1.97 3.56
100% 2.85 3.11 5.75 1.96 2.16 3.93
TABLE IV. Prediction Results in CIELAB E and E
94
for the
HP Series
Series of
constant
yellow ink Maximal Maximal
percentage E
E
n
2
E E
94
E
n
94
2
E
94
0% 2.07 2.34 4.70 0.96 1.06 1.89
25% 2.57 2.85 4.85 1.19 1.28 2.64
50% 2.59 2.90 5.48 1.34 1.52 3.65
75% 3.27 3.49 6.85 1.97 2.13 4.13
100% 3.92 4.28 6.89 2.45 2.68 4.14
Modeling Ink S preading for Color Prediction Vol. 46, No. 3, May/J une 2002 243
r ies. This could be due t o t he st r onger spr ea ding of t he
HP inks.
I n or der t o show how our ink spr ea ding model im-
pr oves color pr edict ion, we consider t he 25 sa mples of
t he ser ies ha ving a 100% cover a ge of yellow ink a nd
wh i ch a r e pr i n t ed on HP Ph ot o Pa per wi t h a n HP
DJ 560C pr int er. The pr edict ion r esult s cor r esponding
t o t his ser ies a r e given in t he la st r ow of Ta ble I I I . The
mea sur ed a nd comput ed spect r a of t he 25 sa mples a r e
con ver t ed in t o CI E-XYZ va lu es t h a t a r e sh own in a
gr a phica l for m in Fig. 10. Figur e 10a shows t he devia -
t ion bet ween mea sur ed a nd comput ed color s t a king ink
spr ea ding int o a ccount ; a nd Fig. 10b shows t he devia -
t ion bet ween mea sur ed a nd comput ed color s wit hout
t a king ink spr ea ding int o a ccount . Wit hout t a king ink
Figure 8. Spr ea ding of HP inks pr int ed on HP pa per a nd spr ea ding of Epson inks pr int ed on Epson pa per. For ea ch of t he 30 non-
equiva lent configur a t ions in a t hr ee-ink pr int , t he spr ea ding coefficient s r = r
i
/r
0
a r e r espect ively given in t he columns ent it led
HP a nd Epson. The second a nd sixt h columns show t he dot configur a t ions. The da shed cir cles indica t e t he loca t ions of neighbor s
t ha t a r e not cover ed wit h ink. The whit e disks cor r espond t o one ink dr op, t he gr ey disks cor r espond t o t wo ink dr ops, a nd t he
bla ck disks cor r espond t o t hr ee ink dr ops. The vect or s indica t e t he or ient a t ion of t he spr ea ding.
spr ea ding int o a ccount , t he pr edict ion er r or is higher
t ha n E = 10 in CI ELAB for sa mples ma de of mor e t ha n
t wo inks, beca use pr int ing ink dr ops one over a not her
significa nt ly incr ea ses ink spr ea ding. The spect r a of
sa mples ma de wit h only one ink a r e well pr edict ed be-
ca use a lmost no ink spr ea ding occur s.
The mea sur ed a nd pr edict ed spect r a of t he cya n a nd
gr een sa mples shown r espect ively in Fig. 1a a nd Fig.
1b a r e given in Fig. 11a a nd Fig. 11b. Sa mple (b) is ma de
of t he sa me cya n la yer a s sa mple (a ) a nd cover ed wit h a
unifor m yellow la yer. I n spit e of t he a bsence of a bsor p-
t ion of t he yellow ink a t 630 nm, t he mea sur ed r eflec-
t ion coefficient of sample (b) is lower t han t hat of sample
(a ) a t t his wa velengt h (compa r e Fig. 11b a nd Fig. 11a ).
This differ ence is expla ined by ink spr ea ding. The cya n
244 J our na l of I ma ging Science a nd Technology
]
]
]
]
|
(
'
'
`
J
J
J
0 0
2
2
1
( )
( )
(9)
Dur ing t he spr ea ding pr ocess, t he a mount of dye T
r ema ins const a nt . We ca n est a blish a r ela t ionship be-
t ween t he ma xima l densit y D
0
a t t he cent er of a dr op
t ha t did not spr ea d, a nd t he densit y D
M
a t t he cent er of
a dr op t ha t spr ea d, by ca lcula t ing T in bot h ca ses.
Wit hin a n isola t ed cir cula r impa ct (r() = r
0
) which
did not spr ea d (D
M
= D
0
), t he t ot a l a mount of dye is:
T D r
2
0 0
2
(10)
Let us now consider a n ink dr op t ha t spr ea d. I n t his
ca se, r() is defined piecewise a s shown in Eq. (1). The
int egr a l in Eq. (9) ca n be wr it t en:
T D
r
d d
i
r
M
i
i
|
(
'
`
J
J
]
]
]
]
|
(
'
'
`
J
J
J
+
1
6
0
2
1
1
( )
( )
(11)
wher e
i
i +
6
1
3
( ) .
nm
nm
246 J our na l of I ma ging Science a nd Technology
+
1
6 2
1
4
[ ( )]
.
(12)
I n t he next st ep, t he following cha nge of va r ia ble is
per for med: t = (3/) ( /6) i. The int egr a l (12) ca n be
wr it t en:
T
D
r r t r r t r dt
M
i
i i i i i
+
[ ]
+ +
12
2 3
1
6
0
1
1
3
1
2
2
( ) ( )
. (13)
By int egr a t ing Eq. (13) wit h r espect t o t he va r ia ble t ,
we get :
T
D
r r r r
M
i
i i i i
+
|
(
'
`
J
J
+ +
12
13
140
1
6
1 1
2
( )
(14)
Fina lly, by combining Eqs. (10) a nd (14) we obt a in
Eq. (3).
Acknowledgment. We would like t o t ha nk Dr. Rit a
Hofma nn fr om I lfor d I ma ging for pr oviding useful in-
for ma t ion on ink jet pa per, a nd t he Swiss Na t iona l Sci-
ence Founda t ion (gr a nt No. 21-54127.98) for suppor t ing
t he pr oject .
References
1. P. Emmel and R. D. Hersch, Towards a Color Prediction Model for
Printed Patches, IEEE Computer Graphics and Applications, 19 (4)
5460 (1999).
2. P. Emmel and R. D. Hersch, A Unified Model for Color Prediction of
Halftoned Prints, J. Imaging Sci. Tech. 44, 351359 (2000).
3. R. Rolleston and R. Balasubramanian, Accuracy of Various Types of
Neugebauer Model, Proceedings of the First IS&T/SID Color Imaging
Conference, IS&T, Springfield, VA, 1993, pp. 3237.
4. P. Emmel, Modles de Prdiction Couleur Appliqus lImpression Jet
dEncre, PhD thesis No. 1857, Ecole Polytechnique Fdrale de
Lausanne (EPFL), Lausanne, Switzerland, 1998, p. 114, http://
diwww.epfl.ch/w3lsp/publications/colour/thesis-emmel.html (in French).
5. C. J. Rosenberg, Measurement-based evaluation of a printer dot model
for halftone algorithm tone correction, J. Electronic Imaging 2(3), 205
212 (1993).
6. G. Plya, Kombinatorische Anzahlbestimmungen fr Gruppen, Graphen
und chemische Verbindungen, Acta Mathematica, 68, 145254 (1937).
7. G. Plya and R. C. Read, Combinatorial Enumeration of Groups, Graphs
and Chemical Compounds, Springer, New York, 1987.
8. R. A. Brualdi, Introductory Combinatorics, Third Edition, Prentice Hall,
London, 1999, pp. 546586.
9. S. Lang, Algebra, Third Edition, Addison-Wesley, New York, 1995, p.
7.
10. Ref. 9, p. 30.
11. R. A. Brualdi, Introductory Combinatorics, Third Edition, Prentice Hall,
London, 1999, p. 575.
12. S. Wang, Algorithm-Independent Color Calibration for Digital Halftoning,
Proceedings of the Fourth IS&T/SID Color Imaging Conference, IS&T,
Springfield, VA, 1996, pp. 7577.
13. S. Wang and C. Hains, US Patent No. 5,748,330 (1998).
14. J. Sawada, Fast Algorithms to Generate Restricted Classes of Strings
Under Rotation, PhD Thesis, University of Victoria, Canada, 2000.
15. http://www.epson.com/
16. http://www.hp.com/
17. P. Emmel and R. D. Hersch, Spectral Color Prediction Model for a Trans-
parent Fluorescent Ink on Paper, Proceedings of the 6th IS&T/SID Color
Imaging Conference, IS&T, Springfield, VA, 1998, pp. 116122.
18. G. Rogers, A Generalized Clapper-Yule Model of Halftone Reflectance,
Color Res. Appl. 25, (6) 402407 (2000).