C

OMPUTER

G

RAPHICS

Proceedings, Annual Conference Series, 1999 SIGGRAPH 99

1

Abstract

This contribution introduces the basic techniques for digital facial
engraving, which imitates traditional copperplate engraving.
Inspired by traditional techniques, we rst establish a set of basic
rules thanks to which separate engraving layers are built on the top
of the original photo. Separate layers are merged according to sim-
ple merging rules and according to range shift/scale masks spe-
cially introduced for this purpose. We illustrate the introduced
technique by a set of black/white and color engravings, showing
different features such as engraving-specic image enhancements,
mixing different regular engraving lines with mezzotint, irregular
perturbations of engraving lines etc. We introduce the notion of
engraving style which comprises a set of separate engraving layers
together with a set of associated range shift/scale masks. The
engraving style helps to port the look and feel of one engraving to
another. Once different libraries of pre-dened mappable engrav-
ing styles and an appropriate user interface are added to the basic
system, producing a decent gravure starting from a simple digital
photo will be a matter of seconds. The engraving technique
described in this contribution opens new perspectives for digital
art, adding unprecedented power and precision to the engravers
work.

Keywords:

photorealistic rendering, nonphotorealistic render-
ing, halftoning, dithering, digital engraving

1. Rationale

Engraving is among the most important traditional graphical tech-
niques. It rst appeared in the fteenth century as an illustrative
support for budding book-printing, but very quickly became an art
in its own right, thanks to its specic expressive power. Actually,
four main classes of engraving are used by artists: letterpress or
relief printing, intaglio or in-hollow printing, silk screen process
and lithography, with several different techniques in each class.
The history of printmaking was punctuated by prosperous periods
of techniques which later declined for various reasons. Facial
engraving is one such example. Extremely popular in seventeenth
and eighteenth centuries, when photography did not exist, this
wonderful art became almost unused, due to the extreme technical
demands that it made on the engraver. Professional copperplate
engravers are rare today, and the cost of true engravings is simply
too prohibitive to be used in everyday printing. At the same time,
traditional facial engraving has no doubt very specic appeal: its
neat, sharp appearance distinguishes it advantageously from pho-
tos. To appreciate the graphical impact of engravings its enough to
compare the engraved portraits in the Wall Street Journal with por-
traits in other newspapers produced with traditional impersonal
screening.
Does it mean that this enjoyable art is condemned to disappear for
purely economical reasons? We dont think so. We do believe that
computer graphics can transform traditional engraving into a digi-
tal art. Already in the past, considerable effort has been made to
transform traditional pen-and-ink illustrations into digital form
[24], [22], to make digital watercolors [3], various line art draw-
ings [5], [6], [7], [8], [9], [15], [21], woodcut imitations [18], and
expressive painting [13], [17].

Fig. 1

Traditional copperplate engravings: an enlargement of a
modern post stamp (left) and an enlargement of a modern
banknote (right).
In the present contribution, we shall try to establish the basis for
digital engraving, and more specically, for facial engraving. Our
goal is very precise: starting from a digital photo of a person, to be
able to reproduce it faithfully, relying on the technical achieve-
ments and techniques that traditional engravers used in the past.
The resulting digital engraving should be visually pleasant, and the
person must be recognizable. We limit our study to one particular
type of facial engraving: copperplate engraving. Although the
terms etching, mezzotint, aquatint and others designate dif-
ferent traditional techniques, we shall sometimes use these terms
as well.
Many authors who tried to produce digital gravures in the past
were too inuenced by their scientic background: they tried to put
engraving layers or similar lineart stuctures on 3D parametric sur-
faces directly attached to the 3D objects (see for example [7], [8],
[9], [16]). Notwithstanding the extreme complexity of such an
approach, it leads to results of a relative value for facial engraving:
simple geometrical objects look too striped, while more sophisti-
cated objects like human face are too complicated to be manipu-
lated easily.
Instead, we tried to reuse the graphical techniques at the disposal
of traditional artists who did not know anything about parametric
surfaces (see Fig. 1). Visibly, the main rule for creating facial
engravings was: the directions of the engraving lines should some-
how follow facial features. Somehow means: just loosely related to
large facial surfaces. Long curves traversing several features were
very welcome (see for example [2], [14]). This rough gure-hug-
ging layered engraving, combined with cross-etching, forms the
basis for our approach.
Technically, the micro-photography of the copperplate engraving
process shown in Fig. 2a gives us an important insight. The width
of the furrows produced by the tip of the graver will inuence the
engraving line width. The furrows have a specic triangular shape
clearly visible in Fig. 2a (please note: they are concave, not con-
vex). Based on this observation, we could build a virtual univer-

Digital Facial Engraving

Victor Ostromoukhov

1

Ecole Polytechnique Fdrale de Lausanne (EPFL), Switzerland

1. http://lspwww.epfl.ch/~victor

2

SIGGRAPH 99, Los Angeles, CA, August 8-13, 1999

sal copperplate as shown in Fig. 2b. This universal copperplate
can be cut at different heights thus producing furrows of different
widths. By putting ink into the furrows we imitate the true copper-
plate engraving printing process. But at the same time the process
described here is nothing other than conventional dithering, well-
known in computer graphics [23]. Simply, the term of virtual uni-
versal copperplate stands for the threshold matrix (threshold lev-
els corresponding to the height of our universal copperplate),
while cutting and inking stands for comparison between the input
signal level and the current threshold value, thus producing a black
or white output signal. The analogy is so perfect that the art of dig-
ital copperplate engraving may be resumed as the art of building
appropriate threshold structures looking like the universal cop-
perplate in Fig. 2b. Once this threshold structure is built, the ren-
dering may be done using conventional dithering software.
In section 2, we establish the basic rules for building separate lay-
ers containing elementary engraving lines, as well as the rules for
merging separate layers together in order to produce the resulting
threshold structure. In section 3, we show how the proposed tech-
nique may be applied to black and white engraving. In section 4,
simple color extensions are proposed. The perspectives for future
work are discussed in section 5. Finally, we draw some conclusions
in section 6.

Fig. 2

(a) A micro-photography of the gravers tip making a fur-
row in the copperplate showing a shaving lifted by the bevel.
(b) Virtual universal copperplate cut at different heights,
producing furrows of different widths. Putting ink into the
furrows imitates the true copperplate engraving printing pro-
cess.

2. Basic rules

Our goal is presently to develop a technique for building separate
engraving layers, to transform them in order to follow the desired
directions and nally to superimpose these layers, thus forming
various cross-etching and smooth transitions between different
parts of the artwork rendered by different engraving layers.
In this section we develop a simple and straightforward technique
based on the most regular support for any line art: a sequence of
equidistant straight lines dened on a given region. We call such a
sequence the basic engraving layer. After appropriate morphing,
the basic engraving layer roughly corresponds to the main etching
technique used in traditional engraving. Later, we shall see how
this relatively simple technique may be extended to imitate more
sophisticated etching techniques such as irregular lines or mez-
zotint.

2.1 Building separate layers

Let us suppose that the basic (non-transformed) engraving layer is
dened on a unit square in the

uv

coordinate system, as shown in
Fig. 3. The layer is built of a sequence of threshold structures made
up of uniformly spaced waves, as shown in Fig. 3b. The cross-sec-
tion of each wave has a simple saw shape, directly inspired by the
shapes of the furrows in Fig. 2a. The directions of the waves are
not necessarily parallel to axes

u

and

v

.

Fig. 3

Parametric grid dened on a unit square in parametric
space

uv

is transformed into morphed parametric grid inside
the patch

T

, in image space

xy

(upper row). This transforma-
tion maps Basic Engraving Layer onto Transformed Layer
(lower row).
The basic engraving layer can easily be transformed into a warped
layer (Fig. 3). There are many ways to perform such a transforma-
tion (see for example [11], [12], or [25]). We have chosen a very
simple and intuitive way to dene the transformation by means of

Coons patches

[4]. The main advantage of this construction con-
sists in the fact that the border curves

U

1

(u),

U

2

(u),

V

1

(v)

and

V

2

(v)

, which delimit the Coons patch, can easily be created on top
of the source photo to be rendered by the engraving, using popular
powerful tools such as Adobe Illustrator. As we want the engraving
lines to somehow follow the directions of the features in the origi-
nal image, we naturally build the border curves taking into account
the borders of the features such as nose, eyes, cheeks, lips etc. The
process of building the border curves



becomes more clear by
observing a concrete example of building an engraving style shown
in the top row of Fig. 5.
The border curves are built of an arbitrary number of straight line
and/or Bzier curve segments (two segments for the curves

U

1

(u),
V

1

(v), V

2

(v),

and three segments for the curve

U

2

(u)

in Fig. 3).
Each curve has to be re-parametrized in order to preserve the uni-
formity of the curves Euclidean length when the parameter uni-
formly walks in the range [

0..1

] (see for example [10], section 9.4).
The re-parametrization permits a smooth and uniform interpolation
between the curves built of a different number of segments, of dif-
ferent lengths.
We suppose that our parametric curves

U

1

(u),

U

2

(u),

V

1

(v)

and

V

2

(v)

form the interior of the closed quadrilateral patch

T

:

U

1

(0)=

V

1

(0)

,

U

1

(1)=

V

2

(0)

,

U

2

(0)=

V

1

(1)

and

U

2

(1)=

V

2

(1)

. Any
point

P

(u,v)=(x,y)

inside the patch

T

can be dened as a function
of parameters

u

and

v

(after re-parametrization), by applying the
linear interpolation between the curves

U

1

(u)

and

U

2

(u),

taking
into account the correction terms

L

(v)

and

R

(v)

due to deviations
by left and right curves

V

1

(v)

and

V

2

(v):

A uniform grid dened in the parametric space

uv

is transformed
using the transformation (1) into a warped grid, as shown in Fig. 3.
Any basic engraving layer dened in the parametric space is
accordingly transformed into a warped engraving layer in the
image space.
(a)
(b)
Universal copperplate
Inking Inking Inking
Cutting Cutting Cutting

u
v
x
y
U
1
(u)
U
2
(u)
V
2
(v)
V
1
(v)

v
u
0 1
1
Patch T
uv: Parametric Space xy : Image Space
(a)
(b)
P u v , ( ) 1 v ( ) U
1
u ( ) v U
2
u ( ) 1 u ( ) L v ( ) u R v ( ) + + + =
L v ( ) V
1
v ( ) 1 v ( ) V
1
0 ( ) v V
1
1 ( ) =
R v ( ) V
2
v ( ) 1 v ( ) V
2
0 ( ) v V
2
1 ( ) =
where 0 u v , 1
1 ( )

3

C

OMPUTER

G

RAPHICS

Proceedings, Annual Conference Series, 1999

2.2 Superimposition of separate layers

Now, our task is to establish basic rules for the superimposition of
several transformed engraving layers, each of which is a simple
matrix of threshold values. For the sake of simplicity, we consider
that the threshold values are in the range between 0 and 1; one
entry of this threshold matrix corresponds to one pixel of the
resulting image. We assume that there is a one-to-one correspon-
dence between the matrix of threshold values representing the
engraving layer and the resulting image. No geometrical transfor-
mation is performed at this stage, only the threshold values may be
affected.
The superimposition of several layers is performed sequentially,
one layer after another. For this reason, it is important to dene a
set of basic rules for superimposing

two

layers, the extension to
several layers being straightforward. Each engraving layer, before
the superimposition, may undergo two range transformations: it
may be scaled (range scale) and raised or lowered (range shift):
where range scale values

S(x,y)

and range shift values

D(x,y)

are
two matrices of the same dimensions as the matrix of threshold
values

T(x,y)

which forms the transformed engraving layer. We
build the range scale and range shift matrices using popular image-
manipulation tools such as Adobe Photoshop on top of the feature
borders in the original image. A trimming operation may be needed
when the resulting threshold value

T(x,y)

goes beyond the range
[

0..1

]. Trimming may be performed either after every layer super-
imposition operation, or at the very end of the whole sequence of
superimpositions.
As we see, superimposing engraving layers consists in consecu-
tively merging the current layer (CL) into the resulting layer (RL).
Once merged, the current engraving layer disappears as an inde-
pendent entity. The merging is performed according to the merging
mode. Table 1 enumerates some merging modes, among the most
important ones. This list in not exhaustive: additional modes may
be added if needed.
Fig. 4 illustrates the use of merging modes. The sample image con-
tains two parts: a uniform gray ramp and four at patches whose
respective intensities are 1/8, 3/8, 5/8 and 7/8. It may be noticed
that the copy smaller and bigger modes are by far the most
useful for engraving purposes. In fact, the copy mode serves to
initialize the resulting engraving layer for the very rst merging
operation. The smaller mode produces cross-etching which is
very close to traditional cross-etching known in the art. Please note
that in the dark area this mode does not produce continuous lines.
The bigger mode is complementary to the smaller mode: it
produces continuous lines in the dark areas, and discontinuous
ones in highlights. Finally, as can be observed in the bottom line of
Fig. 4, a judicious combination of appropriate merging mode with
individual layer range shift may produce very interesting technical
effects: one particular layer may become apparent only in a desired
subrange of gray.

2.3 Equilibration

One may notice that the tone reproduction curve of two superim-
posed layers is no longer linear, even if both layers forming the
superposition have a linear curve reproduction behavior. This
means that when several layers are superimposed, the resulting
engraving may appear locally darker or lighter than what is
expected. In order to cope with this phenomenon, we have to equil-
ibrate the resulting threshold structure: when the dithering process
is applied, this modied threshold structure must produce visually
uniform gray for a uniform input signal of any intensity, indepen-
dently of the number of layers and the superposition rules that have
been used. A simple histogram equalization will not work because
of its global nature. Instead we need local model-based histogram
equalization, taking into account the dot gain of the printing engine
and the characteristics of the human visual system. This dramati-
cally improves the quality of the result because the engraving lines
are often very thin, and their visual impact should be carefully
taken into account. The equilibration process has been described in
detail in another publication [19]. Let us outline it in a few words.

Fig. 4 Merging modes. Left column: the threshold structure
obtained by superimposing threshold structures L1 and L2
using different merging modes. Right column: a sample
image (the topmost image) rendered using the threshold struc-
tures shown in the left column. Used merging modes were:
row a: copy L1, row b: copy L2, row c: L1 smaller L2,
row d: L1 bigger L2, row e: L1 multiply L2, row f:
(L1 scaled 1/2) add (L2 scaled 1/2), row g: L1 smaller
(L2 raised 3/16).
The threshold matrix to be equilibrated is corrected separately, for
different input signals (at uniform surfaces). For each input level,
standard dithering is performed in order to obtain the bitmap to be
sent to the printing device that we modelize. After applying dot-
gain correction and low-pass ltering which simulates the human
visual system, we get the model approximation of the printed and
perceived surface corresponding to the uniform input signal of a
given intensity. The response of the model is usually non-uniform.
According to the local discrepancies between the input and model-
based output, we locally modify the threshold matrix, then redo the
whole cycle of the output model-based simulation described
before. After a few iterations, we obtain the threshold matrix which
produces a reasonably uniform output. Various parameters of the
model are adjusted using the measurements of real test prints.
Once all correction terms for the threshold matrix have been calcu-
lated for all input intensities, the resulting corrected threshold
matrix may be calculated. For performance reasons, we calculate
the correction terms only for a few (16) input intensity levels
spread uniformly throughout the whole intensity range, the rest
being linearly interpolated.
Table 1: Merging (superimposition) modes
merging mode description
.. copy CL T
RL
(x,y) = T
CL
(x,y) * S(x,y) + D(x,y)
.. smaller CL T
RL
(x,y) = MIN(T
RL
(x,y), T
CL
(x,y) * S(x,y) + D(x,y))
.. bigger CL T
RL
(x,y) = MAX(T
RL
(x,y), T
CL
(x,y) * S(x,y) + D(x,y))
.. multiply CL T
RL
(x,y) = T
RL
(x,y) * (T
CL
(x,y) * S(x,y) + D(x,y))
.. add CL T
RL
(x,y) = T
RL
(x,y) + (T
CL
(x,y) * S(x,y) + D(x,y))
T' x y , ( ) T x y , ( ) S x y , ( ) D x y , ( ) + =
Sample Image
Transformed Layers
Resulting Engravings
(a)
(b)
(c)
(d)
(e)
(f)
(g)
4
SIGGRAPH 99, Los Angeles, CA, August 8-13, 1999
Fig. 5 The process of building the engraving style. Upper row: sequence of gure-hugging parametric grids. Middle row: the corresponding
range shift masks. Lower row: the succession of resulting engraving layers during the superimposition process. The merging rules are
indicated on the top of each layer.
Range shift matrices (masks) for successive engraving layers
Parametric grids for successive layers, on the top of the original image
Succession of resulting engraving layers

copy L1 L1 smaller L2 (L1 smaller L2) smaller L3
.. smaller L4
layer L1 layer L2 layer L3 layer L4 layer L5
((L1 smaller L2) smaller L3)
.. smaller L4)
(((L1 smaller L2) smaller L3)
.. smaller L5
The results of our equilibration technique are very satisfactory for
the digital engraving presented in this article, as well as for other
dithering techniques.
3. Black and white facial engraving
Now, let us illustrate the techniques described in the previous sec-
tion showing an example of black and white engraving of the head
of Michelangelos Giuliano de Medici. Five separate engraving
layers for various parts of the face have been created (see Fig. 5,
upper row). The borders of the patches are arranged in such a way
that the patches grids loosely follow the key features of the image:
the nose line, the cheek prole, the eye shape etc.
The middle row of Fig. 5 shows the range shift matrices associated
with the corresponding engraving layers. Here we use the follow-
ing convention: mid-gray corresponds to a zero range shift, white
corresponds to a full-range raise (D(x,y)=+1) and black corre-
sponds to a full-range lowering (D(x,y)=-1), and the gradations
between these three states mean intermediate range shifts. It may
be noticed that the pixels whose range shift values in a given layer
are +1 (white in our convention) do not participate in building the
resulting engraving layer. In such a way, the range shift matrices
act as transfer masks between the current and the resulting engrav-
ing layers. Smoothness of gradation between the areas where shift
D(x,y)=0 and D(x,y)=1 determines the nature of the fusion
between several layers: abrupt boundaries dene neat junctions
between layers whereas smooth boundaries determine a very pro-
gressive fusion between the layers. In our example, the boundaries
of the area where D(x,y)=0 in layer L2 (around the right eye) are
relatively abrupt; the corresponding engraving shows a pretty neat
junction between layers L1 and L2, as can be seen in Fig. 6b. On
the contrary, the junction between layers L1 and L3 in the area of
the left eyebrow is more progressive - and the resulting engraving
shows some overlapping between these layers (Fig. 6b). Its
according to the artists taste that the degree of smoothness
between the layers may be determined.
In the building process shown in Fig. 5 we used only the copy
and smaller merging rules between successive layers. For all lay-
ers, the scale values S(x,y)=1 (i.e. no scaling).
Fig. 6 shows the resulting engraving achieved using the engraving
style shown in Fig. 5 (the only difference between the engraving
layers shown in Fig. 5 and Fig. 6 consists in dividing by two all
etching frequencies in the former, for the sake of visibility). It may
appear quite surprising that a relatively simple technique described
here produces such a decent result.
Furthermore, the visual quality of the engraving may be improved
by applying more sophisticated techniques, and especially various
cross-etchings, as shown in Fig. 7. The principle of cross-etching
shown in Fig. 7a and enlarged in Fig. 7b is very simple: from the
same patch Twe generate two engraving layers: in the rst one,
the engraving crests are oriented following the curvilinear coordi-
nate u, and in the second one - following the curvilinear coordinate
v.
5
COMPUTER GRAPHICS Proceedings, Annual Conference Series, 1999
Fig. 6 The engraving produced using the engraving style shown
in Fig. 5.
Fig. 7c shows that the bigger rule for superimposition of the lay-
ers may play an additional role.
3.1 Mixing etching and mezzotint
Fig. 7 illustrates another expressive tool traditionally used by some
engravers: mixing regular etching and mezzotint. In this example,
the entire layer L5 has been replaced by a specially designed mez-
zotint engraving layer. Enlargement in Fig. 7d shows that our simu-
lated mezzotint is a relatively good match for the traditional
aquatint texture shown in micro-photography in Fig. 7e. We imple-
mented our mezzotint following the description in [20], slightly
modied. Let us note the specically warm appearance of such an
engraving, as well as the acute contrast between regular and sto-
chastic parts. Traditional engravers in the past often made the
most of the additional expressive power of such juxtapositions.
3.2 Real photos
Compared to the photographs of sculptures, real-people photos
may present some additional difculties: the contrast of the key
features such as the prole line, the nose, lips and eye contours
may be insufcient. Real people may have some particular traits
and features that one may wish to hide or stress. Typically, the
wrinkles in female portraits often become almost invisible,
whereas the vigorous lines on the male portraits are sometimes
stressed.
Also, special features such as glasses, moustaches, earrings etc.
may require special technical attention. Real photos may need an
additional pre-processing phase which would perform traditional
cosmetic arrangements, contrast and edge enhancements, as well
as engraving-specic enhancement techniques.
3.3 Engraving-Specic Enhancement
Engraving offers a large set of expressive tools for visual contrast
enhancement. We have already mentioned the effect of mixing
mezzotint and regular etching. Further examples are shown in Fig.
8. The contrast between the glasses and the face is achieved by an
abrupt change of the direction and frequency of the etching. As we
explained before, such an effect can very easily be obtained by
designing appropriate engraving layers, and by abrupt borders in
the range shift masks.
Another tool for stressing particular features like a contour line or a
small detail is illustrated in Fig. 8c. Here, a small additional layer
has been added in order to accentuate the nasolabial fold. It may be
noticed that additional engraving strokes parallel to the feature did
not modify the impression of the gray level of this particular area:
our equilibration process made the other lines locally thinner. On
the contrary, the feature itself (nasolabial fold) appears much more
contrasted.
Fig. 7 The engraving produced using the engraving style shown
in Fig. 5 enriched by the cross-hashing and the mezzotint. The
upper enlargement shows cross-etching using smaller rule,
whereas the lower enlargement shows another engraving
which uses both smaller and bigger rules. The enlarge-
ment in the middle of the upper row is compared with the
micro-photography of a real aquatint copperplate.
bigger and smaller rules
smaller rule
micro-
photography
6
SIGGRAPH 99, Los Angeles, CA, August 8-13, 1999
Fig. 8 This engraving shows various enhancement techniques: an abrupt change of the orientation and frequency of the etching (in glasses),
or an additional layer for sharper appearance of the nasolabial fold in (c).
(a)
(b)
(c)
Yet another, more subtle enhancement technique consists in irregu-
lar perturbation of regular engraving layers. The simplest case is
visible in the background layer of the bottom-left color engraving
in Fig. 9, where the engraving lines have been unevenly disturbed.
This effect provides a feeling of tension, especially when com-
pared with the mezzotint background of the bottom-right color
engraving in Fig. 9.
3.4 Engraving Style
The set of layers which form the engraving together with range
shift and scale matrices form what may be called the engraving
style. The usefulness of this notion consists in the possibility of
using the same engraving layers for different photos of the same
person, as well as for producing engravings of different persons,
with minimal modications. Engravings shown in Fig. 6 and Fig. 8
represent two different (although quite close) engraving styles. On
the contrary, engravings in Fig. 6 and Fig. 9 use almost identical
styles for the face itself, for the same person (two lower engravings
in Fig. 9), as well as for different persons (two left engravings in
Fig. 9). Only minor modications were needed in order to adapt
the engraving style from one person to another.
The engraving styles in Fig. 6 and Fig. 9 have been adjusted to the
photos manually. Another possibility of making such an adjust-
ment automatically or semi-automatically would be to use the tech-
nique of feature-based image metamorphosis described in [1]. In
fact, the resulting engraving layer for person A is a two-dimen-
sional threshold structure, rich enough to be subjected to some
morphing which would map the features of person A to those of
person B. This work is under way.
4. Color engraving
Traditional engraving is essentially black and white art. This does
not remove the temptation to experiment with colors. Color raster
images like digital photos are usually stored in Red-Green-Blue
(RGB) or Cyan-Magenta-Yellow-Black (CMYK) representation.
For the sake of simplicity let us consider the rst case.
The simplest color scheme would be to use the same resulting
engraving layer for all three RGB components: it is known as in-
phase color printing. It certainly works but the achieved visual
effect is not very different from black and white gravure.
One may expect a more advanced visual effect when the different
color layers use different orientations. A simple but efcient
scheme has been experimented: for the Red and Blue color planes,
we use the engraving layers with line orientation along the para-
metric axis u, whereas for the Green color plane, we use the same
engraving layers but with line orientation along the parametric axis
v. This produces very nice color cross-etching visible in the
engravings in Fig. 9. This technique may be called engraving by
orthogonal (in parametric space) color lines.
Yet another possibility is to mix normal regular engraving for some
color planes with the mezzotint for the others. The enlargement in
the middle of Fig. 9 shows a detail where all three color engraving
techniques enumerated here are shown in the same picture: the blu-
ish background is done using in-phase engraving with the same
slightly disturbed lines for all three RGB color planes; the boys
forehead has been produced with orthogonal color lines, whereas
the boys hair is a mixture of regular engraving (Red and Blue
color planes) with mezzotint (Green color plane). We hope that
more advanced color mixing / texture mixing schemes may intro-
duce unexpected, beautiful visual effects.
5. Future Work
The technique presented in this contribution may be extended in
several directions.
The rst direction, and probably the most important one, is to build
libraries of pre-dened mappable engraving styles. As we men-
tioned before, the feature-based image metamorphosis presented
by Beier and Neely, or a similar algorithm may work: in our case,
as well as in [1] we have to deal with 2D structures only. With such
libraries of pre-dened styles, one will be able to chose one style
among several proposed, and the whole job of engraving produc-
tion will be a matter of seconds. The engraving styles of the sam-
ples shown in this contribution contain 5 to 10 separate layers,
7
COMPUTER GRAPHICS Proceedings, Annual Conference Series, 1999
whereas more elaborate library styles may contain dozens of sepa-
rate layers. If needed, specic image-dependent features can be
added on top of the basic style.
Another promising direction is experimentation with color engrav-
ing as well as with different engraving textures. The techniques for
color engraving shown in this contribution are relatively straight-
forward extensions of black and white engraving. Although the
results shown in Fig. 6 to Fig. 9 may be considered satisfactory, we
see a good opportunity to go beyond the imitation of existing tech-
niques. Computers offer us unlimited computational power, and
one may experiment with very sophisticated graphical techniques,
inaccessible to traditional engravers for purely technical reasons.
Finally a substantial effort to provide an appropriate user interface
has to be made. In the current implementation we have developed a
very rudimentary user interface. To be attractive for a nal user
who, we hope, will be an artist rather then a programmer, the
engraving system should have ergonomics comparable to that of
the best lineart products such as Adobe Illustrator. Feature-based
layer construction guidance, inter-layer constraint control, simple
and intuitive mapping of the pre-dened styles - all these function
should be incorporated into a powerful GUI.
6. Conclusions
We have presented a very simple technique for producing digital
engravings. The proposed system is based on the analogy between
the universal copperplate which imitates the true copperplate
engraving technique and conventional dithering. The art of digital
copperplate engraving may be resumed as the art of building
appropriate threshold structures.
We have developed the basic technique for building separate
engraving layers (threshold structures) which roughly follow the
features of the original image, as well as the rules for merging
them together. The resulting threshold structure is equilibrated in
such a way that it generates a visually uniform output for a uniform
input signal of any intensity. Applied on an input digital photo,
using a standard dithering algorithm, such a threshold structure
generates a reasonably faithful reproduction, which imitates tradi-
tional engraving. Several enhancement techniques, specic to
engraving, have been proposed. The important notion of engraving
style which comprises a set of separate engraving layers together
with range shift and scale masks has been introduced. Engraving
styles make it easier to adapt the look and feel of an engraving of
person A to an engraving of person B. Finally, a simple color
extension has been proposed.
With this contribution, we consider that the main goal that we set
ourselves starting with a digital photo, to be able to make a digi-
tal engraving of reasonable quality in a reasonably short time has
been achieved. In perspective, additional features such as libraries
of pre-dened mappable engraving styles, special color and texture
effects and an appropriate user interface will certainly make such a
system attractive and usable by most of graphists.
7. ACKNOWLEDGEMENTS
I would like to express my gratitude to several people who partici-
pated in discussions touching the subject of this article: Roger Her-
sch, Nicolas Rudaz, Isaac Amidror, David Salesin, Frdric Pighin,
Catrin Petersen, Gershon Elber, among many others. Special
thanks come to Monique Lazega, a professional engraver, who ini-
tiated me to the traditional engraving art. Also, many thanks to my
kids, Bella, David and Michael, for lending their nice faces.
Finally, many thanks to anonymous SIGGRAPH reviewers for
their helpful suggestions and comments.
REFERENCES
[1] T. Beier and S. Neely, Feature-based image metamorphosis. In
Computer Graphics (SIGGRAPH'92 Proceedings), 26(2), pp.
35-42, 1992.
[2] F. Brunner, A Handbook of Graphic Reproduction Process,
Hasting House Publ., New York, 1984.
[3] C. Curtis, S. Anderson, J. Seims, K. Fleischer, D. H. Salesin.
Computer-Generated Watercolor, Proceedings of SIGGRAPH
97, in Computer Graphics Proceedings, Annual Conference
Series, pp. 421-430, 1997.
[4] S. A. Coons, Surfaces for Computer Aided Design of Space
Forms, MIT Project TR-41, MIT, Cambidge, MA, 1967.
[5] O. Deussen, J. Hamel, A. Raab, S. Schlechtweg, T. Strothotte.
An illustration technique using intersections and skeletons,
Graphics Interface 99, Kingston, Ontario,1999 (to appear) .
[6] D. Dooley and M. F. Cohen. Automatic illustration of 3D geo-
metric models: Lines. Computer Graphics, Vol. 24, No. 2, pp
77-82, 1990.
[7] G. Elber. Line Illustrations in Computer Graphics. The Visual
Computer, Vol. 11(6), pp 290-296, 1995.
[8] G. Elber. Line Art Rendering via a Coverage of Isoparametric
Curves, IEEE Transactions on Visualization and Computer
Graphics, Vol. 1, No 3, pp 231-239, September 1995.
[9] G. Elber., Line Art Illustrations of Parametric and Implicit
Forms, IEEE Transactions on Visualization and Computer
Graphics, Vol 4, No 1, pp. 71-81, 1998.
[10] G. Farin, Curves and Surfaces for Computer Aided Geometric
Design, Academic Press, 1990.
[11] J. D. Foley, A. van Dam, S. K. Feiner, J. F. Hughes, Computer
Graphics, Principles and Practice, Second Edition, Addison-
Wisley Publ., 1990.
[12] J. Gomes, L. Darsa, B. Costa, L. Velho, Warping And Mor-
phing Of Graphical Objects, Morgan Kaufman Press, 1998.
[13] P. Haeberli, Paint by numbers: Abstract image representation.
in Computer Graphics (SIGGRAPH'90 Proceedings), Vol. 24,
pp. 207-214, 1990.
[14] W. M. Ivins, Jr., How Prints Look, John Murray Publ., Lon-
don, 1988.
[15] J. Lansdown, S. Schoeld, Expressive Rendering: A Review
of Nonphotorealistic Techniques, IEEE Computer Graphics
and Applications, Vol. 15(3), pp. 29-37, May 1995.
[16] W. Leister, Computer Generated Copper Plates, Computer
Graphics Forum, Vol.13(1), pp. 69-77, 1994.
[17] B.J.Meier, Painterly Rendering for Animation. Proceedings of
SIGGRAPH 96, in Computer Graphics Proceedings, Annual
Conference Series, pp. 477-484, 1996.
[18] S. Mizuno, M. Okada and J. Toriwaki. Virtual Sculpting and
Virtual Woodcut Printing, in Visual Computer, 10, pp. 39-51,
1998.
[19] V. Ostromoukhov, R. D. Hersch, Multicolor and Artistic Dith-
ering, Proceedings of SIGGRAPH'99, In ACM Computer
Graphics, Annual Conference Series, 1999.
[20] V. Ostromoukhov, R. D. Hersch, Stochastic Clustered-Dot
Dithering. In Proc. SPIE Vol. 3648, Color Imagig: Device-
Independent Color, Color Hardcopy, and Graphic Arts IV, pp.
496-505, 1999.
[21] Y. Pnueli, A. M. Bruckstein, Digidrer - a digital engraving
system, The Visual Computer, Vol. 10, pp. 277-292, 1994.
[22] M. P. Salisbury, M. T. Wong, J. F. Hughes, and D. H. Salesin.
Orientable Textures for Image-Based Pen-and-Ink Illustra-
tion, Proceedings of SIGGRAPH 97, in Computer Graphics
Proceedings, Annual Conference Series, pp. 401-406, 1997.
[23] R. Ulichney, Digital Halftoning, The MIT Press, Cambridge,
Mass., 1987.
[24] G. Winkenbach, D. H. Salesin. Rendering parametric surfaces
in pen and ink. Proceedings of SIGGRAPH 96, in Computer
Graphics Proceedings, Annual Conference Series, pp. 469-
476, 1996.
[25] G. Wolberg, Digital Image Warping, IEEE Computer Society
Press, 1990.
8
SIGGRAPH 99, Los Angeles, CA, August 8-13, 1999
Fig. 9 Examples of color engraving.