See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/332794094

Topological Modeling for Vector Graphics

Article  in  IEEE Computer Graphics and Applications · May 2019

DOI: 10.1109/MCG.2019.2891277

CITATIONS READS

0 16

2 authors, including:

Boris Dalstein
VGC Software
4 PUBLICATIONS   48 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Vector graphics and animation View project

All content following this page was uploaded by Boris Dalstein on 08 May 2019.

The user has requested enhancement of the downloaded file.

 Dissertation Impact

Topological Modeling
for Vector Graphics
Boris Dalstein
VGC Software

Editor: Jim Foley [foley@cc.gatech.edu]

Editor’s note:
Boris Dalstein received the 2017 Alain Fournier Award for the best Canadian
Computer Graphics Dissertation.

Abstract—Vector graphics refers to the use of geometrical primitives, such as Be
zier

curves, to represent digital images. It is becoming increasingly popular among graphic
designers who need to deliver resolution-independent content that looks sharp across all
types of desktop and mobile devices, or need to support user interactivity and animation.
Unfortunately, most vector graphics tools today have many limitations, such as the
inability to represent shapes sharing a common edge. In my doctoral dissertation, we
address this issue by developing a novel data structure, called the vector graphics
complex, which supports fundamental topological modeling operations for vector graphics
illustrations. We also extend this data structure to animation, allowing features of a
connected drawing to merge, split, appear, or disappear at desired times via keyframes
that introduce the desired topological change. The resulting space-time continuous
complex directly captures the time-varying topological structure, and allows features to
be readily edited in both space and time in a way that reflects the intent of the drawing.

& MY PHD DISSERTATION is about representing
the connectivity between two-dimensional (2-D)
shapes. For example, a user might draw a rectan-
gle, then draw another rectangle, and finally
decide that the two rectangles should share
a common edge, thus, effectively “gluing together”
Digital Object Identifier 10.1109/MCG.2019.2891277
Date of current version 24 April 2019.
two of the existing edges (see Figure 1). While
this seems simple enough, it turns out to be
a challenging problem in the general case, for
counterintuitive reasons, but before going into
more details, let me tell you a relevant story.
No more than a few months ago, while I was
curiously wandering through the numerous
aisles of SIGGRAPH 2018’s exhibition floor, what a
surprise! I noticed a reproduction of Sutherland’s

0272-1716 ß 2019 IEEE Published by the IEEE Computer Society IEEE Computer Graphics and Applications
86
Figure 1. Simple topological operation on 2-D
vector graphics shapes: Gluing two rectangles along
a shared edge.
Sketchpad program,14 developed in 1963, and
widely recognized as one of the first interactive
graphical programs. Barely anyone else seemed
to notice it, or care much about it, but I was truly
ecstatic and I immediately started to play with it
for a few minutes. Why the excitement? I had grad-
uated not long before, and Sutherland’s PhD dis-
sertation about Sketchpad is cited in the first
sentence of my own dissertation.
The reason for telling you this story is that my
dissertation can be seen as a direct follow-up of
Sutherland’s brilliant and highly impactful disser-
tation from 1963. And as of today, I am still baffled
that my dissertation had not been written at least
30 years ago by somebody else. Indeed, apart
from a few other topological data structures
developed in the 1980s, I am essentially not build-
ing on any other work more recent than that. Simi-
larly to how most of SIGGRAPH’s attendees failed
to notice the beautiful reproduction of Sketchpad,
the computer graphics community seems to have
failed to notice that such a deceptively simple
problem—representing the connectivity between
2-D shapes—was still largely unsolved.
1963–2018: A STEP BACKWARD?
One of the core functionalities of Sketchpad
was to allow users to simultaneously edit three
or more incident lines by simply moving the
point where they meet (see Figure 2, top). In the
abstract of his dissertation, Sutherland writes:
Sketchpad stores explicit information
about the topology of a drawing. If the user
moves one vertex of a polygon, both adja-
cent sides will be moved. If the user moves
a symbol, all lines attached to that symbol
will automatically move to stay attached
to it. The topological connections of the
May/June 2019
Figure 2. Sketchpad allows the explicit representation of n-way
joins. Today’s most popular vector graphics editors and file
formats do not.
drawing are automatically indicated by the
user as he sketches.
Therefore, already back in 1963, Sutherland
understood the value of explicitly storing topolog-
ical information about a drawing, such as the inci-
dence relationship between lines: it allows users
to interact with the drawing in intuitive ways
which would otherwise not be possible. In fact,
not only did Sutherland pioneer vector graphics,
but he also pioneered topological modeling.
However, something surprising happened.
Fast-forward to 2018, the leading commercial
vector graphics editor in the industry, Adobe
Illustrator CC, poorly supports this feature.
The leading open-source vector graphics editor,
Inkscape, does not support this feature at all.
The leading and widely adopted open file format
for vector graphics, W3C SVG 1.1, does not sup-
port this feature at all. Despite how obviously
desirable this feature is, despite being described
in the abstract of one of the most well-known
PhD dissertation in computer graphics history,
despite the apparent simplicity of implementing
this feature, and despite more than 50 years of
development of the field, this feature is largely
absent in today’s vector graphics landscape.
What went wrong? How did this happen? One
can only speculate. However, it may be a hint
that there are some nonobvious, hidden issues
that make implementing this feature harder
than it looks: some intrinsic properties of vector
graphics that went in the way and some funda-
mental problems that 50 years of research failed
to identify and solve.
87
 Dissertation Impact
Figure 3. Unlike our data structure, no previous vector graphics
representations support overlapping shapes with shared edges.
Figure 4. Allowing shapes to overlap, and allowing them to be
glued together along edges may result in nonplanar topologies. For
example, the object on the right is nonplanar: it is an orientable
surface of genus 1, homeomorphic to a torus with a missing disk.
Somehow, a fascinating subfield of computer
graphics, topological modeling for vector graphics,
failed to materialize in the 1970s to tackle these
problems. My dissertation is a long overdue
analysis of these problems, and a first attempt to
solve a few of them.
MIXING OVERLAPPING
WITH TOPOLOGY
A major limitation of Sketchpad, however, is
that it does not support “color filling.” In other
words, Sketchpad only allows the representation
of 1-D lines, while today’s vector graphics
editors also support the representation of 2-D
shapes, which is arguably more important than
supporting topological modeling. In this sense,
the period 1963–2018 is in fact not a step back-
ward for vector graphics, but rather a step in a
different direction, adding some features (color
88
filling), but dropping others (topological model-
ing). Still, it is surprising that so little research
has been done to try and reconcile the two, espe-
cially when topological modeling is common-
place within the 3-D modeling world. Is not 2-D
easier than 3-D? We will answer this question
later in this paper, but let us first get an intuition
of the challenges to overcome.
The vast majority of today’s vector graphics
editors use an internal representation similar to
that of the W3C SVG 1.1 standard. Topologically
speaking, this representation is extremely sim-
ple: it is a sequence of independent curves
assigned attribute values such as “stroke-width,”
“stroke-color,” and “fill-color,” which is enough
to represent 2-D shapes. These shapes are
allowed to overlap each others (see Figure 3,
left), but they cannot share a common edge. It is
possible to create two shapes that appear to
have a common edge, but in fact no topological
information is explicitly stored, and the
“common edge” is in fact two unrelated edges
that just happen to have the same geometry.
Less commonly, other vector graphics editors
(or subtools such as Live Paint groups in Illustra-
tor) are based on planar maps.3,2 Planar maps are
able to represent a partition of the 2-D plane into
nonoverlapping regions, which is stored using a
halfedge data structure. They are able to explic-
itly store the topological connections between
two shapes (see Figure 3, middle), but by defini-
tion they cannot represent overlapping shapes. In
practice, this means that most vector graphics
tools are not compatible with this representation,
and any geometric edit results in an expensive
recomputation of the planar map, potentially los-
ing previously authored attributes. This severely
impedes artistic freedom.
Unlike these previous approaches, our new
data structure6 is able to support both overlap-
ping and topological modeling (see Figure 3, right).
The reason why this is more complicated than it
seems is that by allowing shapes to overlap, and
also allowing to “glue” these shapes together
along edges, we may in the general case end
up with nonplanar topologies (see Figure 4), or
even nonorientable topologies (see Figure 5),
which are a special case of nonplanar topologies.
By definition, such nonplanar topologies cannot
be embedded in R2 (only an immersion can be
IEEE Computer Graphics and Applications
defined), which breaks many assumptions com-
monly done in vector graphics, and brings a few
complications detailed in my dissertation.
DATA STRUCTURE
Our data structure (see Figure 6), called vector
graphics complex (VGC), is a topological structure
made of four types of cell: vertices, open edges,
closed edges, and faces. A vertex is simply defined
as a 2-D point. An open edge is defined as a 2-D
directed curve that starts and ends at some verti-
ces, possibly equal. A closed edge is defined as a
2-D directed closed curve, with no start or end
vertex at all. Finally, a face is defined as a 2-D
closed region delimited by vertices and edges,
ordered in cycles. A cycle is defined as either: a
closed sequence of consecutive open edges; or a
single closed edge; or a single vertex.
Typically, one of the cycles represents
the outer boundary of the face, and the other
cycles represent inner holes, possibly degener-
ate (missing curves/points). However, because
the cycles of a given face may intersect, the
distinction between outer boundary and inner
holes is in general ill-defined. For this reason,
the VGC does not explicitly store whether a
cycle is an outer boundary or an inner hole.
Cycles made of consecutive open edges may
use the same edge e any number of times, and
each of these edge-uses independently specifies
whether e is traversed in its intrinsic direction, or
in the opposite direction. We use the term half-
edge to refer to an edge with such specified direc-
tion, and we respectively denote by e and e
1 the
two possible directions. This allows us to denote
cycles using a convenient multiplicative notation,
such as g ¼ e2 e
1
1
5 e6 e6 e7 . Cycles made of a single
closed edge e may use it multiple times, but all
uses must be with the same direction, i.e., either
g ¼ en or g ¼ e
n . This represents a cycle that
circles around the same closed edge multiple
times. Finally, cycles made of a single vertex v are
called Steiner cycles and denoted g ¼ v.
COMPARISON WITH 3-D MODELING
Our data structure is a boundary-based
representation (B-rep), and has many similari-
ties with existing data structures used for 3-D
topological modeling. In fact, many 3-D modeling
May/June 2019
Figure 5. Top: Constructing a topological Mo €bius strip via simple
user interactions. Although users only manipulate an immersion in
R2 , the represented surface would be nonorientable if embedded
in R3 , that is, it would only have one side. Bottom: Comparison
with an annulus. Note how the Mo €bius strip and the annulus are
both composed of two vertices, three edges, and one face, but
the unique cycle of the face is different: in the case of the Mo €bius
strip, e2 is used two times in the same direction, while in the case
of the annulus, e2 is used once in each direction. Constructing
topological Mo €bius strips in R2 is not possible if edges are not
allowed to overlap.
applications only represent the surface of 3-D
shapes, that is, 2-D objects. The fact that in our
case, we immerse these 2-D objects in R2 rather
than R3 has little influence on the underlying
topological model. However, our specific use
case targets less technical users with higher
expectations in terms of artistic freedom, which
requires us to support a few special cases that
are typically not supported in 3-D modeling
applications.
For example, as we detail in the dissertation,
not only do we need to be able to represent non-
planar topologies, but also nonorientable topolo-
gies. This excludes the use of the popular halfedge
and winged-edge data structures,4 which can only
represent orientable surfaces. Also, 2-D artists
often use a variety of techniques to depict their
subjects, often including thin features drawn as
a single line, such as the whiskers of a cat. This
results in frequent use of nonmanifold objects
(see Figure 7), which cannot be represented by
the quad-edge data structure,10 for example.
89
 Dissertation Impact
Figure 6. Example C++ implementation of the VGC. In addition to the minimal information required, this implementation
also stores the star of each cell c, which is the set of cells c0 whose boundary contains c.
Potential data structures supporting mixed-
dimensional nonmanifold topologies include
simplicial complexes (= triangulations), the
Figure 7. Example of nonmanifold topologies supported by our
data structures. In the case of surfaces with boundary, “manifold”
means “everywhere locally homeomorphic to R2 or R  ½0; 1Þ”,
that is, no tie-vertices (bottom-left), no 1-D features (bottom-
middle, and right), and every edge must have either one or two
incident faces.
90
radial-edge data structure,15 or generalized
maps.12 Unfortunately, they all have shortcom-
ings for our use case, such as the need to decom-
pose the object into more entities than an artist
would like to. For example, 2-D artists often like
to be able to represent and manipulate closed
curves as one single entity, without the need to
arbitrarily have to split the curve with a vertex.
It would be possible to use these other struc-
tures as underlying representations hidden by the
user interface, but this does not remove the
need to define a higher level layer of abstraction.
In other words, VGCs could be implemented
in terms of triangulations, but the need for a
user-friendly “VGC interface” still remains, and
implementing this interface in terms of triangula-
tions is not necessarily helpful. A relevant
data structure with almost all our desiderata
IEEE Computer Graphics and Applications
(e.g., able to represent a closed curve as one
entity) is the selective geometric complex,13 but
unfortunately it can only represent embeddings,
while we also require to represent immersions.
At the end of the day, we chose a representation
most similar to Weiler’s radial-edge structure,15
but removing some features that we do not need,
and adding some others that we do need.
2-D ANIMATION WITH TIME-
VARYING TOPOLOGY
As a follow-up to this paper, we extended this
data structure to support animation.7 In particu-
lar, we were interested to be able to represent
animations of shapes whose topology continu-
ously changes over time. Although we published
this work a year later, it was in fact our original
motivation, and I had been working on this prob-
lem since my Master’s.
Although relatively rare in the world of 3-D
animation, time-varying topology is the norm
rather than the exception in 2-D hand-drawn
animation. Indeed, since each keyframe is manu-
ally drawn by hand, consecutive keyframes tend
not to use the same number of strokes or
the same connections between these strokes
(see Figure 8). This makes it really hard to auto-
mate the generation of inbetween frames, that is,
to automatically interpolate between two hand-
drawn keyframes. This problem has been identi-
fied early on by Ed Catmull in 1978,5 and remains
largely unsolved today. Attempted solutions
are either pixel based and result in blurring
artefacts,1 or are limited to cases that could
be solved via layering11 (that is, hidden
strokes are automatically reconstructed), or
only provide automation for the subset of the
keyframes with consistent topology, and let
the artist manually draw the remaining subset
with inconsistent topology.16
None of the approaches really had an effec-
tive mathematical representation to describe
the topological changes going from one image to
another. Such a representation seemed like an
obvious prerequisite to solve the inbetweening
problem, therefore we decided to focus on this.
Our idea was to represent a 2-D animation as
a 3-D space-time topological complex, as illus-
trated in Figure 9.
May/June 2019
Figure 8. Topology of these two keyframes is not the same: how
to interpolate between them?
TOPOLOGICAL KEYFRAMING
In order to get an intuition behind this space-
time structure, consider a simpler example: ani-
mating a point P that splits into two points Q
and R. To specify such animation, one needs to
represent:
 An animated position P ðtÞ for t 2 ½t0 ; t1 ,
where t0 is some time before the split, and t1
is when the split occurs.
 Animated positions QðtÞ and RðtÞ for t 2
½t1 ; t2 , where t2 is some time after the split.
The first observation is that we must have
P ðt1 Þ ¼ Qðt1 Þ ¼ Rðt1 Þ. To avoid redundancy, we
store this value only once as a pair v1 ¼ ðt1 ; p1 Þ
which we call a key value, and which can be seen
as the temporal equivalent of storing only once a
shared edge between two shapes. Other key values
we need are v0 ¼ ðt0 ; P ðt0 ÞÞ, v2 ¼ ðt2 ; Qðt2 ÞÞ, and
v3 ¼ ðt2 ; Rðt2 ÞÞ, respectively, storing the initial
position of P , and the final positions of Q and R.
Once we have defined this set of key values
V ¼ fv0 ; v1 ; v2 ; v3 g, we simply need to specify that
v0 must be interpolated into v1 , and then that v1
must be interpolated into both v2 and v3 . This can
be done by interpreting V as the nodes of a graph
G, where ðv0 ; v1 Þ, ðv1 ; v2 Þ, and ðv1 ; v3 Þ would be the
edges of this graph. We call this paradigm
“topological keyframing”: it contrasts with the
more conventional paradigm (which we call
“sequential keyframing”), where a fixed number of
animation variables are defined and then animated
independently, making it impossible for the num-
ber of degrees of freedom to change over time. See
Figure 10, for a comparison of these paradigms.
Our proposed space-time topological com-
plex is an extension of this idea to represent not
only points that split and merge, but also curves
and 2-D shapes that split and merge. We refer to
the paper for more details, but in particular, one
91
 Dissertation Impact
Figure 9. Time-continuous 2-D animation of a bird represented as a 3-D space-time topological complex. Top: Space-
time view. Bottom: output animation.
can notice that within this paradigm, an ani-
mated 2-D shape is in fact a volume in space-
time, and representing volumes, especially with
nonmanifold topology, is notoriously hard.
PRESENT AND FUTURE IMPACT
Our approach has been implemented in
the open-source prototype VPaint (https://www.
vpaint.org), which has been downloaded around
15 000 times as of this writing, and we have
received spontaneous feedback from a few dozen
92
users (most likely the most enthusiastic users, so
this should be taken with a grain of salt: we did
not conduct any formal user study). Overall, the
feedback was extremely positive, with users being
excited by the potential of the technology both for
static images and animation. However, the current
lack in our prototype of many features commonly
found in other editors (e.g., brush styles, textures,
grids, guides, compositing, grouping, cloning,
alignment tools, curve tangents editing, text fea-
tures, color management, better import–export
options, better timeline editing operations) makes
IEEE Computer Graphics and Applications
Figure 10. Left: The existing keyframing paradigm, defining an animation as ordered sequences of key values. Right: Our
more general approach, where key values are unordered but labeled, and inbetween values specify which ones to interpolate.
it frustrating to use for anything other than toy
examples. Also, while topological operations are
generally extremely fast by nature (just a few bits
to update), we did not optimize at all rendering
and it is currently a bottleneck in VPaint, making
the sketching experience subpar past a few hun-
dred strokes. That being said, we have no reason
to believe that rendering a VGC should be notice-
ably slower than rendering any other vector
graphics representation, once optimized using
comparable techniques.
Several researchers have leveraged the
expressiveness of the VGC data structure for
their use cases. Garcia et al.9 developed a
method to assist geologists in finding out and
describing a sequence of geological events (e.g.,
the formation of a mountain range), where the
topological connections between geological
layers are stored using the VGC. This is quite an
unexpected use case, but our data structure is in
fact eminently suitable for it. More related to our
intended use case, Entem et al.8 proposed an
approach to convert a given contour drawing of
organic shapes (animals, humans, etc.) into a
layered structure with added semantics, using
the VGC as their output format.
Unfortunately, there is yet to be seen any
work using or extending our space-time data
structure for 2-D animation. There are a few par-
allel attempts to tackle the problem of automatic
inbetweening,17; 18 but none using a method that
May/June 2019
explicitly tracks time-continuous topological
changes as we do, that is, either they are pixel
based and subject to blurring artefacts, or they
do not support topological changes. Vector-
based approaches could certainly benefit from
using the VGC.
Not surprisingly, current 2-D drawing pro-
grams have not been updated to use the VGC,
as this new drawing paradigm represents quite a
significant shift, changing at its core the internal
representation of illustrations and animations.
This makes it almost impossible for existing
tools to adopt it without rewriting most of their
backend, and it is typically impossible to imple-
ment the VGC as a plugin since the core
representation of a program is generally not
customizable. This is obviously a big barrier
for adoption.
Also, it is important to note that hand-drawn
animation requires a lot of artistic skills, precision,
and patience, and that it is very unusual for aca-
demic researchers to have the time to build a user
interface polished enough for artists to be able to
do any meaningful testing. Also, finding skilled tra-
ditional animators in the first place is quite diffi-
cult. As a consequence of all these factors,
progress in this field is typically quite slow.
In the long term, it is entirely possible that
commercial tools will incorporate VGC concepts.
As a strict superset of the representations
currently in use, it enables useful drawing
93
 Dissertation Impact
paradigms without sacrificing any other fea-
tures, and it is quite simple to implement if doing
it from scratch (replacing an existing architec-
ture may be more complicated). If deemed too
confusing for beginner users, the additional
features such as allowing the representation of
shared edges may even be disabled by default,
and only made available on an opt-in basis.
However, there still remain a few details to
figure out before such structures are production-
ready. For example, now that it is possible to rep-
resent three strokes incident at a common vertex,
it is necessary to be able to render this junction
appropriately. This is not necessarily trivial if the
strokes have different widths or colors: what
should the geometry of the junction look like?
Should it be a round or a Miter join? How should
the color or brush style transition from one stroke
to the others? All of these are currently open
questions, but we believe are reasonably answer-
able, and it is likely that different tools choose dif-
ferent solutions, providing artists with different
controls on how to style these junctions.
Other open-ended questions are about subtle
choices in the user interface: how should the clas-
sical concept of “groups” interact with the topol-
ogy? Should we allow shapes in different groups
to share common edges? How about the concept
of “layers”? There are also questions related to
the selection mechanism: should selecting a face
automatically select its bounding edges? How
about deleting a face, should it preserve its
boundary by default? All these questions have no
equivalent in the representations currently in
use, where a face and its boundary are typically
represented as one single curve entity with a
“stroke-color” and “fill-color” attributes.
Exploring these questions is one of the
goals of my new startup company VGC Software
(https://www.vgc.io). By developing a full-
featured professional vector graphics editor
based on the VGC data structure, I hope to
attract more users, whose interactions with the
software might provide more insight on the
relevance of this data structure, and its long-
term prospect. By keeping the software open-
source and embedding a Python console, I
also hope that it will make research in this
field more accessible, avoiding the need for
graduate students to implement from scratch
94
a selection mechanism, a rendering engine,
brush styles, layers, and others. Indeed, while
3-D software typically has a lot of modularity
to test out techniques as plugins, it is much
less common in 2-D software, slowing down
academic research.
Finally, as far as 2-D animation is concerned,
we are only scratching the surface of what can
be possible. Surprisingly, this field seems much
less mature than 3-D animation, with only very
basic tools commercially available compared to
what is available for 3-D animation. With our
research, we now have a data structure to repre-
sent the topological events which are so com-
mon in 2-D hand-drawn animation, but we still
have no good tools to automatically generate
such topology from scanned drawings, for exam-
ple. More generally, how to create, manipulate,
and visualize this structure in a way that is use-
ful for users is still mostly open, and an exciting
future direction to explore.
& REFERENCES
1. M. Alexa, D. Cohen-Or, and D. Levin, “As-rigid-as-
possible shape interpolation,” in Proc. SIGGRAPH,
2000, pp. 157–164.
2. P. Asente, M. Schuster, and T. Pettit, “Dynamic planar
map illustration,” ACM Trans. Graph., vol. 39, no. 3,
pp. 30:1–30:10, 2007.
3. P. Baudelaire and M. Gangnet, “Planar maps: An
interaction paradigm for graphic design,” in Proc. CHI,
1989, pp. 313–318.
4. B. G. Baumgart, “Winged edge polyhedron
representation,” Tech. Rep. STAN-CS-72-320, DTIC
Document, Stanford Univ., Stanford, CA, USA, 1972.
5. E. Catmull, “The problems of computer-assisted
animation,” SIGGRAPH Comput. Graph., vol. 12, no. 3,
pp. 348–353, 1978.
6. B. Dalstein, R. Ronfard, and M. van de Panne, “Vector
graphics complexes,” ACM Trans. Graph., vol. 33, no. 4,
pp. 133:1–133:12, 2014.
7. B. Dalstein, R. Ronfard, and M. van de Panne, “Vector
graphics animation with time-varying topology,” ACM
Trans. Graph., vol. 34, no. 4, pp. 145:1–145:12,
Jul. 2015.
8. E. Entem, A. Dev Parakkat, M.-P. Cani, and L. Barthe,
“Structuring and layering contour drawings of organic
shapes,” in Proc. Joint Symp. Comput. Aesthetics
Sketch-Based Interfaces Modeling Non-Photorealistic
IEEE Computer Graphics and Applications
 Animation Rendering, Vol. 18, Vic., Canada,
Aug. 2018, pp. 1–14.
9. M. Garcia, M.-P. Cani, R. Ronfard, C. Gout, and
C. Perrenoud, “Automatic generation of geological
stories from a single sketch,” in Proc. Expressive, Vic.,
Canada, Aug. 2018, pp. 17–19.
10. L. Guibas and J. Stolfi, “Primitives for the manipulation
of general subdivisions and the computation of
Voronoi,” ACM Trans. Graph., vol. 4, no. 2, pp. 74–123,
Apr. 1985.
11. A. Kort, “Computer aided inbetweening,” in Proc. 2nd
Int. Symp. Non-Photorealistic Animation Rendering,
2002, pp. 125–132.
12. P. Lienhardt, “Subdivisions of n-dimensional
spaces and n-dimensional generalized maps,” in
Proc. 5th Annu. Symp. Comput. Geometry, 1989,
pp. 228–236.
13. J. Rossignac and M. O’Connor, “SGC: A dimension-
independent model for pointsets with internal
structures and incomplete boundaries,” Res. Rep. IBM
T. J. Watson Research Center, Yorktown Heights, NY,
USA, 1989.
May/June 2019
View publication stats
14. I. E. Sutherland, “Sketchpad, a Man-Machine Graphical
Communication System,” Ph.D. dissertation,
Massachusetts Inst. Technol., Cambridge, MA, USA,
1963.
15. K. Weiler, “Topological structures for geometric
modeling,” Ph.D. dissertation, Rensselaer Polytechnic
Inst., Troy, NY, USA, 1986.
16. B. Whited, G. Noris, M. Simmons, R. Sumner, M. Gross,
and J. Rossignac, “BetweenIT: An interactive tool for
tight inbetweening,” Comput. Graph. Forum, vol. 29,
no. 2, pp. 605–614, 2010.
17. J. Xing, L.-Y. Wei, T. Shiratori, and K. Yatani,
“Autocomplete hand-drawn animations,” ACM
Trans. Graph., vol. 34, no. 6, pp. 169:1–169, Oct.
2015.
18. W. Yang, H.-S. Seah, Q. Chen, H.-Z. Liew, and
D. S
ykora, “FTP-SC: Fuzzy topology preserving stroke
correspondence,” Comput. Graph. Forum, vol. 37,
no. 8, pp. 125–135, 2018.
Contact department editor Jim Foley at foley@
cc.gatech.edu.
95