a survey of pen-and-ink illustration in non-photorealistic ...€¦ · terminologies in hand-drawn...
TRANSCRIPT
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 1 of 25
A Survey of Pen-and-Ink Illustration in Non-photorealistic
Rendering
CS361
Computer Science Department
GWU
Jia Huang
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 2 of 25
Table of Contents
A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering ............................. 1Introduction ......................................................................................................................... 3Background ......................................................................................................................... 4
Terminologies in hand-drawn pen illustration ................................................................ 4Principles......................................................................................................................... 5Classification................................................................................................................... 6
Object-based Approaches.................................................................................................... 8Algorithm1 ...................................................................................................................... 8Algorithm2 ...................................................................................................................... 8Algorithm3 .................................................................................................................... 10Algorithm4 .................................................................................................................... 11Algorithm5 .................................................................................................................... 13Algorithm6 .................................................................................................................... 14
Image-based Approaches .................................................................................................. 16Algorithm7 .................................................................................................................... 16Algorithm8 .................................................................................................................... 17Algorithm9 .................................................................................................................... 19
Application........................................................................................................................ 19Future Research................................................................................................................. 22References ...................................................................................................................... 23
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 3 of 25
Abstract
This paper presents a survey of the work done in the pen-and-ink illustration field of non-
photorealistic rendering. Non-photorealistic rendering(NPR), as a relatively new area is
getting more and more attention from the computer graphics community. Pen-and-Ink
illustration, as one of the styles of NPR, received much research in the recent fifteen
years. In this paper, important terminologies and principles are first introduced. Then the
general pen-and-ink illustration techniques are categorized and examined with certain
focus. Afterwards one important application field is discussed. The paper also suggests
important areas for future research.
Introduction
Computer graphics research has focused on photorealistic rendering, which attempts to
create images of physical scenes with ever-increasing realism, that looks “just like the
real world.” However it is not as effective and efficient as non-photorealistic rendering in
a lot of situations. Non-photorealistic rendering (NPR) is any technique that produces
images of simulated 3d world in a style other than realism[1]. There are many styles of
NPR, for example, water color[2], impressionistic[3,4,32], pen-and-ink illustration,
engraving[5], etching[1], etc. Because of the broad spectrum of the area, the focus of the
paper is put on the pen-and-ink illustration style.
Pen-and-ink is an extremely limited medium, allowing only individual
monochromatic strokes of the pen. However beautiful pen-and-ink illustrations
incorporating a wealth of textures, tones, and styles[18] can be created. Indeed, because
of their simplicity and economy, there are a lot of applications and advantages of this
illustration technique. With only a simple form, image creators are able to use expressive
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 4 of 25
abstraction to emphasize the area of interest and focus the viewer’s attention without
being forced to depicting every detail. Therefore it is very effective and efficient for
purpose of illustrating and expressing(i.e. “effectively conveying the meaning or
feeling[28]), especially if the underlying models are too complex to render realistically.
Second, in architectural design and industrial design, stylized illustrations are often more
appropriate, especially in the initial design phase, than photorealistic pictures and CAD
because they give the impression of approximation and incompleteness and therefore are
more stimulating[30,27,25]. Third, because the files created using this method usually
consume less storage, they are reproduced more easily and transmitted more quickly and
especially convenient for laser printers. Forth, the illustrations made in this style are
blended well with texts, using the same ink as texts. Because of these characteristics and
advantages, pen-and-ink illustrations are widely used in textbooks, repair manuals,
advertising and many other forms of media.
First we will discuss some fundamental terminologies, essential principles in pen-
and-ink illustration and a general classification of the general algorithms described here
will also be presented.
Background
Terminologies in hand-drawn pen illustration
For further discussion and instruction, interested reader should consult Guptil[6], a
classic comprehensible text on pen and ink illustration.
There are two different kinds of marks or strokes in pen-and-ink illustration:
outlines, hatching. Outlines are the external boundaries and internal edges, used to define
shapes. Outlines are exceptional in that they may be long and individually significant.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 5 of 25
The choice of whether or not to use outlines is largely an aesthetic one. Hatching is used
to indicate shading but should also follow the surface curvature.
Pen-and-Ink limitation in being monochromatic is overcome to some extent
through the use of tone, style and texture. Tone is the darkness of a section of a drawing.
The perceived gray level or tone in an illustration depends largely on how dense the
strokes are in a region. Style is the brokenness of a line, for example solid line, dashed
line, dotted line. Texture is a tactile impression of a surface as rough, sandy, smooth.[7]
Texture is the collective result of many pen strokes, each individual stroke is not critical
and need not be drawn precisely. Indeed a certain amount of irregularities in each stroke
is desirable to keep the resulting texture from appearing mechanical. The most commonly
used textures include: hatching formed by roughly parallel lines; cross-hatching formed
by overlapped hatching in several directions; and stippling formed by small dots or very
short lines.
Principles
These are some fundamental principles of illustrating in pen and ink, useful for purely
computer generated illustrations. Interested readers are referred to [6] , [8] and [15].
Strokes must look natural, not mechanical, the thickness of a line should vary
along its length, wavy lines are a good way to indicate that a drawing is schematic and
not yet completely resolved. It is not necessary to depict each individual tone accurately,
however presenting the correct arrangement of tones among adjacent regions is essential,
to disambiguate objects it is sometimes important to “ force tone” by enhancing contrast
or inventing shadows. And the character of strokes and outlines is important for
conveying texture, as well as geometry as lighting, e.g. straight lines are good for “glass”.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 6 of 25
S. Strassmann[9] “Hairy brush” proposed the “path-and-stroke” metaphor to emulate
brush using four different objects: brush, stroke, dip and paper to get a variety of
strokes[25]. S. Hsu and I. Lee[12] extended the metaphor by using general objects like
textures, images and recursively defined fractals that are drawn along a given path.
Haeberili[4] also focused on manipulating individual stroke by inspecting a collection of
attributes of the stroke, including location, color, size, direction and shape, and a path is
defined and physically simulated brush is used to generate the stroke.
Classification
There are three distinct types of input for stylized depiction processes[1]:
(1) 3D scenes (described in terms of geometry, color, lighting, etc.) for rendering
(2) images for processing
(3) brush strokes from a user ( like the input to a paint system)
Depending on the scene input(1,2), all the techniques presented here can be
classified into two classes, object-based and image-based[10]. Image-based systems
produce their illustrations directly from grayscale images. In object-based category, there
are two different kinds of image rendering algorithms, the image-centered and the scene-
centered algorithms[11]. Scene-centered algorithms project un-occluded objects onto the
image. Such techniques are qualified for polygonal scenes, while for other object kinds
special treatment is necessary, especially for free form surfaces. Image-centered
algorithms typically have a post-processing phase. Depending on (3), these techniques
can be categorized along the axis from interactive to fully automatic[1]. Below is a
classification of the general techniques that we will examine afterwards, represented by
tags.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 7 of 25
Object-based
Object-centered Image-centered
Image-based
automatic Algorithm4,
Algorithm5,
Algorithm6
Algorithm2,
Algorithm3
Algorithm9
interactive Algorithm1 Algorithm7,
Algorithm8
The advantage of geometry-based(object-based) systems is because of the
availability of the 3D geometry and viewing information they can produce illustrations
whose strokes not only convey the tone and texture of the surfaces in the scene, but also
convey the 3D forms of the surfaces by placing strokes along the natural contours of
surfaces. Existing image-based systems, on the other hand, until now have been able to
convey 3D information only by having a user draw individual strokes or specify
directions for orienting particular strokes. Whereas the ability to generate illustrations
with an image-based system offers several advantages. First, using an image-based
system greatly reduces the tasks of geometric modeling and of specifying surface
reflectance properties. Second, an image-based system provides the flexibility of using
any type of physical photograph, computer-generated image, or arbitrary scalar, vector or
tensor field as input, allowing visualization of data that is not necessarily physical in
nature. Finally, image-based systems offer more direct user control by providing the
ability to modify tone, texture, or stroke orientation[10]. General techniques of pen-and-
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 8 of 25
ink illustration are discussed in the following sections with an emphasis on the
advantages and disadvantages of the system, important steps and key features of the
specific system, the generation and placement of strokes( including rendering of outlines,
strokes, textures and tones) and applications of the algorithm.
Object-based Approaches
Algorithm1
Dooley and Cohen[13] proposed a system to enhance a traditional shading images with
illustration techniques. Their system is able to handle the difficulties arising from
triangulating complex objects when there are many surfaces with common borders
present. Objects in the scene are attached to a set of semantic attributes that are
interactively determined. The rendering algorithm works with illustration rules that are
applied to the projected lines according to their attributes. They attempt a taxonomy of
line styles and semantics. They showed how line and surface qualities could be
customized by the user to create more effective images. For example, lines can vary
thickness, transparency and style and line attributes are described by a matrix on the
values of “importance”, “line type” and “hiddenness”.
Algorithm2
To produce outlined and contoured drawings instead of shaded image, T. Saito et al. [14]
propose an enhancement technique for 3D shapes that conceptualizes geometric
properties. This is an automatic image-centered approach, because all operations are
realized with 2D image processing operations and with no user interaction. The problems
with this method contain: aliasing and reflected or transparent images can not be
enhanced because each pixel can represent only one surface; inefficiency in both
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 9 of 25
execution time and memory space because all images are preserved as floating point data
in order to avoid digitization errors.
The technique can be divided into 3 separate processes: 1.geometric process based
on geometric factors such as object shapes and camera parameters(projection and hidden
surface removal); 2.physical process based on physical factor such as reflectance, colors,
textures(shading, texture mapping); 3.artificial process, such as outline enhancement. The
authors introduce a special “G-buffer” for geometric information used repetitively by the
algorithms. The contents of the buffer include: object/patch identifier, parametric patch
coordinates, Z-buffer, world coordinates and normal vector for the visible patch. G-buffer
is formed during the geometric process and used by the physical and artificial processes.
The basic enhancement operations, i.e. the drawing of discontinuity lines, contour lines
and curved hatching, are done using G-buffer contents during post-processing. It is
separated from geometric and physical processes.
Edge, Contour and Hatching Drawing
Edge contains profile and internal edge. Profile is the border line of an object on the
screen; internal edge is a line where two faces meet. Profile and internal edge are the 0th
and first order discontinuity of the depth image respectively. Discontinuity can be
extracted with a first order differential operator and discontinuity of the first order
differential of an image can be extracted with a second order differential operator. The
artifact resulted from these operations, such as confusing real discontinuities from rapid
depth change can be corrected using the minimum and maximum of neighboring
differential values. So both profile and internal edges can be drawn.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 10 of 25
Using image processing technique, contour lines are generated as raster data and
consists of homogeneous calculations on each pixel and its neighbors. Curved hatching
rendering of a surface is performed using the contour method above in combination with
a technique that thins out and disappears the contours when they become too dense and
adds lines when the lines become too thin; the purpose is to have an overall periodic set
of lines in pixel space.
Application
This algorithm is commonly used in hand drawn illustrations in industrial design, line
illustration, topographical maps, medical imaging and surface analysis. The method is
also useful for photorealistic rendering.
Piranesi[19] system proposed by Landsown and Schofield also uses NPR to create
illustrations from 3D models. Piranesi uses a standard graphics pipeline to create a 2D
reference image akin to a G-buffer. The user is then allowed to select specific regions of
the image and apply textures that emulate natural media interactively or automatically.
Algorithm3
Leiser[11] presents a technique to emulate cooper-plate rendering, an engraving
technique used for old system of printing. The goal is to render a copper plate, drawing
that consists of lines of varying thickness and of single points for 3d scene. A ray-tracing
approach is used to render curves on free-form objects. An advantage of this approach is
that it easily handles reflections and shadowing because ray tracing is used.
The method uses a kind of volume texturing in connection with image processing
algorithms and is suitable for implementation in a ray tracing algorithm. Shading is done
by regular hatchings in several thickness and distances. Copper plates can be generated
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 11 of 25
from 2D pixel images with filter algorithms and image processing techniques. In the post-
processing step an edge detection algorithm is applied. Edges are sudden changes in the
image domain, that are difficult to determine with an image-centered method. This
technique is able to deal with not only polygonal scene but also freeform surfaces.
Experience shows that this method is especially interesting for illustration in books and
for generating icons on user interface.
Algorithm4
Winkenbach et al.[15] propose an automatic object-centered rendering system. Compared
to [Algorithm2 and Algorithm6] the use of strokes are more expressive. And this method
is resolution-independent while a lot of other computer drawing programs do not scale
well. The limitation is that it can only illustrate polyhedral models and can not be used
for curved surfaces. Also flat-shaded surfaces are assumed.
Main Steps
Firstly, the system computes the visible surfaces and the shadow polygons. Secondly, it
projects the polygons to NDC space to generate certain data structure. Thirdly, each
visible surface is then rendered. The procedural texture attached to each surface is
invoked to generate the strokes conveying the correct texture and tone for the surface.
Then all the strokes are clipped to the visible portion of the surface. Finally the outlines
are drawn by extracting form the structure generated in step2.
Key Features
Their rendering system is a basic graphics pipeline with a few notable changes. Some of
changes include rendering of texture and tone, clipping and outline. Because polygons are
no long scan converted, both texture and tone must be conveyed with hatching. And
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 12 of 25
because the stroke is now distorted or displaced in some way, it is important to allow
strokes to sometimes stray slightly outside of the clipped region. To achieve this, we clip
the straight-line path of our strokes prior to adding in the function for waviness. And
boundary and interior outlines must be drawn in a way that takes into account both the
textures of the surrounded regions and the adjacency information.
Rendering Texture and tone
To render the texture and tone, we use prioritized stroke textures. A prioritized stroke
texture is a set of strokes each with an associated priority. When rendering a prioritized
stroke texture, all of the strokes of highest priority are drawn first; if the rendered tone is
still too light, the next highest priority strokes are added until the proper tone is achieved.
For example for a brick texture, the outlines of the individual brick elements have highest
priority, the strokes for shading individual bricks have medium priority, and the hatching
strokes that go over the entire surface have lowest priority. In the cross-hatching texture,
vertical strokes have priority over horizontal strokes, which have priority over the various
diagonal stroke direction. We express texture with outline. Each stroke texture has
associated with it a boundary outline.
Outline drawing
The interior outlines are used within polygons to suggest shadow directions or to give
view-dependent accents to the stroke texture. And for the sake of principles of pen-and-
ink illustration, we minimize outline by drawing it only if the tones of two neighboring
faces are not sufficiently different for disambiguation. Accented outlines, i.e., thickening
edges is a technique for providing subtle but important cues about the 3d aspects of an
illustrated scene. For example, the interior outlines of each brick in the “brick” stroke
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 13 of 25
texture are drawn according to their relationship with the directin of the light source:
brick edges that cast shadows are rendered with thickened edges while illuminated brick
edges are not drawn at all. In addition to the light source direction, the viewing direction
is another important parameter that should be taken into account when drawing outline
strokes.
Algorithm5
Winkenbach et al.[16] extend the previous method[Algorithm4], and propose one
algorithm that can handle curved surfaces formulated parametrically, such as B-spline
surfaces, NURBS and surfaces of revolution. [Algorithm2] mainly uses image processing
technique in the post-processing stage for outline and hatching generation. While this
method integrates aspects of 2D and 3D rendering. In addition, traditional texture
mapping techniques can be used to extend the range of effects that can be achieved with
pen-and-ink rendering. The biggest limitation is that it deals only with surfaces
possessing a global parameterization. One possible solution is to parameterize such
surfaces.
Generation of Stroke Textures
The progress compared to their last implementation is in the generation of the stroke
texture. Firstly they use a grid of lines, which consists of parallel lines, running in one or
more user-specified directions in the parameter domain, to orientate hatching strokes
along a surface. Secondly, they introduce a technique called “controlled-density
hatching”. This technique allows strokes to gradually disappear in light areas of a surface
or in areas where too many strokes converge together and allows new strokes to gradually
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 14 of 25
come into existence in dark areas or areas in which the existing strokes begin to diverge
too much. Specifically a recursive algorithm is used.
Figure 1. Illustration with Texture Map
Controlled-density hatching allows “fine grain” control over the tone of a pen-and-ink
illustration. With this new capability, we can use traditional texture mapping to vary the
tone on the surface of an object.
Algorithm6
Markosian et al.[17] build a system to deliberately trade accuracy for speed. In contrast
Winkenbach’s pen-and-ink rendering system produces decidedly finer images, but takes
several minutes to do so. This system is very quick and can be easily extended to other
styles. But there is no shadow created.
Key Features
This real-time NPR technique is based on economy of line – the idea that a great deal of
information can be effectively conveyed by very few strokes. This algorithm only renders
silhouettes, certain user-chosen key features and some minimal shading of surface
regions. One obstacle to achieving real-time NPR is the problem of determining
visibility, since a straightforward use of z-buffer may give incorrect results.
So the key idea is rapid identification of silhouette edges using interface
coherence of silhouette edges and fast visibility determination using improvements and
simplification in Appel’s hidden-line algorithm.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 15 of 25
Main Steps
The overall structure of the algorithm is: 1.determine the silhouette curves in the model
2.determine the visibility of silhouette and other feature edges by a modified Appel’s
algorithm 3.render the silhouette and feature edges.
Generating Strokes
World-space polylines to be rendered are first projected into the film plane. Artistic or
expressive strokes are then generated by modifying the resulting 2D polylines. There are
three techniques for generating expressive strokes: drawing the polylines directly with
slight enhancements such as variations in line width or color; adding offsets to the
polylines to define high-resolution “artistically” perturbed strokes; and texture-mapped
strokes which follow the shape of the polyline. In the second case the polylines are
parameterize by arc length. A new parametric curve Q(t) is based on the original
parametric curve p(t) by adding a vector offset v(t) defined in the tangent-normal basis,
i.e. Q(t) =p(t) + vx(t)p’(t) + vy(t)n(t). The third method builds a texture-mapped mesh
using the polyline as a reference spine. Each texture map represents a single brush stroke.
The textures are repeated along the reference spine, approximately preserving its original
aspect ratio.
Placement of Strokes
Shading strokes (particles) are put in world space (not the surface) rather than define
them in screen space. This is the approach used by Meier in her “painterly rendering”[3].
The advantage of this approach is it maintains frame-to-frame coherence. Stroke
directions are defined by the cross product of local surface normal and the ray from the
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 16 of 25
camera to the stroke location and the ray from the camera to the stroke location, so that
stroke line up with silhouette lines.
Applications
Haro[24] implements a non-photorealistic renderer for producing pictures that look like
sketches using this technique. Interested reader may refer to [24] for details.
Image-based Approaches
Algorithm7
Salisbury et al.[18] present an algorithm for rendering subdivision surface models of
complex scenes using an interactive editable particle system, with 2D grayscale images
as a starting point.
Interaction
Strassmann [9] presents interactive simulation of traditional artist tool. Compared to his
approach, rather than focus on individual strokes, this new system tries to directly support
the higher-level cumulative effect that the strokes can achieve: texture, tone and shape.
The interaction between the user and system is high-level in that the user “paints” using
textures and tones, and the computer draws the individual strokes. Exceptions are outlines
that have individual significance; in addition an artist might occasionally need to touch
up fine details of textured work.
Texture generation
Different from Winkenbach al. ‘s approach, a combination of non-procedural and
procedural stroke textures are used. Non-procedural texture is that the textures tiled the
plane and the stroke selected for drawing at a point was the one that happened to pass
through that point. A library of user-defined stored stroke textures is supported, as well as
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 17 of 25
several built-in procedural stroke textures. A stored texture is simply a collection of
strokes. Drawing a texture at a given darkness is a matter of choosing from the collection
a subset that has enough strokes to reach the desired tone. For some textures such as
scribbles, the choice of strokes to draw for a given tonality is not critical. In these cases
the system simply selects strokes from the texture in a random sequence, generating
candidate strokes and testing the tonal effect of candidate strokes. For other textures, the
system supports a predefined priority for each stroke. In addition to modifying individual
strokes, the user can edit collections of strokes, like the “lighting” operation which
incrementally removes strokes.
Outline Drawing
The system allows scanned, rendered or painted images to be used as a reference for tone
and shape. The reference image is used for several purposes: as a visual reference for the
artist, as a tone reference for painting, as a source image from which the outline is
extracted, as a reference for determining stroke and texture orientation.
Algorithm8
P. Salisbury et al. [10] are the first to use orientable textures for image-based pen-and-ink
illustration. This is also an improvement of the previous method[Algorithm7]. This
system is able to render strokes and stroke textures according to a vector field in such a
way that they also produce the proper texture and tone; and estimate tones as new
oriented strokes are progressively applied.
Interaction
Editor allows the user to specify the three components of a layer (tone, direction, and
texture), the system does the tedious work of placing all the strokes.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 18 of 25
Figure 2. The User Input(from left to right) tone, direction and a stroke example
In their previous work, the textures tiled the plane and the stroke selected for
drawing at a point was the one that happened to pass through that point. By contrast, in
this new system the placement of strokes on the final illustration is independent of their
relative position in the texture. Spacing between strokes is instead maintained indirectly
by the rendering system.
Placement of Strokes
Dynamic placement of stroke is import to maintain the density. Dynamic is implemented
by drawing in order of importance, the fraction of its intended darkness that has not yet
been accumulated at that point. Rendering consists of looking for the location with
greatest importance, placing a stroke, update an image that records the importance and
repeat until the importance everywhere is below a termination threshold. The whole
process of matching the illustration to the tone image is recorded in a difference image,
updated after each stroke is drawn, whose value at each pixel is the difference between
the tone image and the blurred version of the illustration. The importance image is
derived from the difference image, its value at each point is the current difference derived
by the initial value of the difference. This algorithm is called Difference Image
Algorithm.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 19 of 25
After knowing where to put the stroke, orient and bend the stroke. Map the stroke
into the direction field so that at every point along its length, the stroke’s new angle
relative to the direction field is the same as the prototype stroke’s angle with respect to
the vertical direction.
Algorithm9
Curtis[26] implements a “loose and sketchy” filter to automatically draw the visible
silhouette edges using image processing and a stochastic, physically-based particle
system. The only input is a depth map of the model and it will be converted into two
images: template image and force field. Template image represents the amount of ink
needed for each pixel. And the force field pushes particles along the silhouette edges.
The particles are generated randomly initialized by the template image. Acceleration of
each particle is based on the force field. Particle generates ink and remove ink during its
traveling, a technique used in [10, 31]
Application
In the following text, one application area of pen-and-ink illustration, the rendering of
trees or complex natural objects, is discussed .The advantage of art-based illustration is
evident in the kind of application where the underlying model is so complex that it is very
time-consuming to model and render while the whole thing can be rendered with a few
strokes which evoke the impression of complexity using pen-and-ink illustration. Trees
are one of those complex objects.
Kowalski et al. [20] suggest an algorithm to render fur, grass, trees etc. by pen-
and-ink illustration. This approach is image-centered and interactive. Their approach is to
generate abstract sketches of trees by using geometric primitives like spheres for defining
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 20 of 25
rough approximations of a tree’s foliage. These primitives were rendered conventionally
to achieve gray-scale images. In the second step, the images were used to procedurally
place graftals – small objects representing leaves or hair – on the surfaces by applying the
“difference image algorithm” proposed earlier by Salisbury[Algorithm8]. Their
algorithm controls the density of hatching strokes in order to match the gray tones of the
target image. And this method enables different textures assigned to different regions,
while in the previous version, [Algorithm6] all the surfaces are assigned with textures
uniformly. It is accomplished by dividing models into one or more regions (patches).
The particle placing is a hybrid of screen and object. To meet the requirement
that graftals appear to stick to surfaces in the scene, graftals are placed in the scene after
converting the 2D screen position to a 3D position on some surface[3].
Figure 4,5. scene rendered without graftal texture(left) and with graftal texture(right)
Fraftals are based on a flat tapering shape by gradually reducing width about a
central spine, which is a planar polyline. After being placed with the DIA, each graftal
determines how to orient and draw itself.
The drawbacks of the algorithm[20] are that each new graftal texture requires a
procedural implementation that included writing code. Also graftals are regenerated in
each frame in a way that leads to excessive introduction and elimination of graftals even
for small changes in camera parameters. Thirdly graftals choose from among a small
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 21 of 25
number of discrete “levels of detail” at which to draw, making transition between levels
noticable.
Markosian et al.[21] propose a new system which overcome these drawbacks by
a more expressive interface, static placement and continuous levels of detail.
The new approach is to use a “static” placement scheme, where graftals are
distributed onto surfaces during the modeling phase. Certain graftals have a multi-
resolution structure, so that a single graftal, seen from larger distance, transforms into
several graftals, when viewed from nearby. Further, these transformations are carried out
in a continuous manner by smoothly varying the shape, size, position and orientation.
Deussen et al.[22] take a very different approach which models detailed tree trunk
and leaves while the goal of the previous approaches is to avoid complex modeling. The
lines drawn are the result of visually combining many drawing primitives instead of
placing graftal objects on some large geometries. A drawback of this approach is that
they potentially have to deal with more input data. The solution to this problem is to
represent a tree at several levels of detail, for example if the current model has too many
leaves a much simpler model can be used instead. The advantage of this method is the
ability to draw both an abstract tree and a specific plant which will be not very different
from its realistic image and the ability to make use of existing tree libraries.
To draw the trunk, the method raised by Markosian in [17][Algorithm6] to render
the outline of the trunk can be used. The skeleton can be crosshatched using “difference
image algorithm” by Salisbury[19][Algorithm7]. The direction of the strokes is similar to
what is used in [17] [Algorithm6] either at random or affected by the normal vector of the
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 22 of 25
stem geometry. Leaves are drawn using abstract drawing primitives. For example, each
leaf can be represented as the outline of a disk.
Future Research
1. Improve the procedural stroke textures and automate further methods for creating
them
2. Incorporate other illustration effects
3. Add more interactive controls to help designing 3d illustrations
4. Render other natural objects
5. Create animation
6. Explore other forms of illustration besides pen-and-ink, including traditional
forms like water color and air brushing
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 23 of 25
References
[1] Craig Reynolds. Stylized Depiction in Computer Graphics. http://www.red3d.com/cwr/npr
[2] Cassidy J. Curtis, Sean E. Anderson, Kurt W. Fleischer, David H. Salesin. In SIGGRAPH ’97
Conference Proceedings.
[3] Barbara J. Mier. Painterly Rendering for Aniamtion. In SIGGRAPH ’96 Conference Proceedings.
[4] Paul Haeberli. Paint by Numbers: Abstract Image Representation. In SIGGRAPH ’90 Conference
Proceedings, Volume 24, Number 4, August 1990.
[5] Victor Ostromoukhov. Digital Facial Engraving. In SIGGRAPH ’99 Conference Proceedings.
[6] Arthur L. Guptil. Rendering in Pen and Ink. Watson-Guptil Publications, New York, 1976.
[7] Tom Jazen. Ten Papers on Non-Photorealistic Rendering. http://www.world.std.com/~tej/pni.html
[8] Frank Lohan. Pen and Ink Techniques. Contemporary Books, inc., Chicago, 1978.
[9] Steve Strassmann. Hairy Brush. In SIGGRAPH ’86 Conference Proceedings, Volume 20, Number 4,
August 1986.
[10] M. Salisbury, M. Wong, J.F. Hughes, and D. Salesin. Orientable textures for image-based pen-and-ink
illustration. In SIGGRAPH ’97 Conference Proceedings.
[11] Wolfgang Leister. Computer Generated Copper Plates. Computer Graphics Forum, Volume 13,
Number 1, March 1994, pp. 69-77, Blackwell, ISSN 0167-7055.
[12] S. Hsu and I. Lee. Drawing and animation using skeletal strokes. In SIGGRAPH ’94 Conference
Proceedings, pages 109-118.
[13] Debra Dooley, Michael F.Cohen. Automatic Illustration of 3D Geometric Models: Lines. In
SIGGRAPH ’90 Conference Proceedings, Volume 24, Number 2, pages 77-82, March 1990.
[14] Takafumi Saito, Tokiichiro Takahashi. Comprehensible Rendering of 3D Shapes. In SIGGRAPH ’90
Conference Proceedings, Volume 24, Number 4, pages 197-206, August 1990.
[15] G. Winkenbach and D. Salesin. Computer-generated pen-and-ink illustration. In SIGGRAPH ’94
Conference Proceedings, pages 91-100.
[16] G. Winkenbach and D. Salesin. Rendering parametric surfaces in pen and ink. In SIGGRAPH ’96
Conference Proceedings, pages 469-476.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 24 of 25
[17] Lee Markosian, Michael A. Kowalski, Sam Trychin, Lubomir Bourdev, Daniel Goldstein, John F.
Hughes. Real-Time Non-Photorealistic Rendering. In SIGGRAPH ’97 Conference Proceedings.
[18] Michael P. Salisbury, Sean E. Anderson, Ronen Barzel, David H. Salesin. Interactive Pen-and-Ink
Illustration. In SIGGRAPH ’94 Conference Proceedings.
[19] Piranesi. http://www.arct.cam.ac.uk/research/cadlab/irender/index.html
[20] Michael A. Kowalski, Lee Markosian, J.D.Northrup, Loring S. Holden, John F. Hughes. Art-based
Rendering of Fur, Grass, and Tree. In SIGGRAPH ’99 Conference Proceedings.
[21] Lee Markosian, Barbara J. Meier, Michael A. Kowalski, Loring S. Holden, J.D. Norhthrip, John F.
Hughes. Art-based Rendering with Continuous Levels of Detail. In NPAR’00 Conference Proceedings.
[22] Oliver Deussen, Thomas Strothotte. Computer-generated Pen-and-ink Illustration of Trees. In
SIGGRAPH ’00 Conference Proceedings.
[23] Thomas Strothotte, Bernhard Preim, Andreas Raab, Jutta Schumann, David R. Forsey. How to Render
Frames and Influence People. In Computer Graphics Forum (13) 3, Proceedings of euroGraphics 1994,
pages 455-466,1994.
[24] Antonio Haro. A nonphotorealistic render. http://www.cc.gatech/edu/~hato/cs7490
[25] Stefan Schlechtweg. Lines and How to Draw Them. http://isgwww.cs.Uni-
Magdeburg.DE/~stefans/pubi/norsigd2.html
[26] Cassidy Curtis. Loose and Sketchy Animation.
http://www.cs.washington.edu/homes/cassidy/loose/sketch.html
[27] Paul Bourke. Computer Sketching. http://www.swin.edu.au/astronomy/pbourke/fractals/sketch/
[28] Matthew Kalplan, Bruce Gooch, Elaine Cohen. Interactive Artistic Rendering. In NPAR’00
Conference Proceedings.
[29] Maic Masuch, Stefan Schlechtweg, Bert Schonwalder. DaLi – Drawing Animated Lines!. In
Proceedings of Simulation and Animtation ’97, S.87-96, SCS Europe, 1997.
[30] Jutta Schumann, Thomas Strothotte, Stefan Laser. Assessing the Effect of Non-Phototealistic
Rendered Images in CAD. http://www.acm.org/sigs/sigchi/chi…dings/papers/Schumann/chi96fi.html
[31] Greg Turk and David Banks. Image-Guided Streamline Placement. In SIGGRAPH ’96 Conference
Proceedings.
Cs361: A Survey of Pen-and-Ink Illustration in Non-photorealistic Rendering Jia Huang
Page 25 of 25
[32] Peter Litwinowicz. Processing Images and Video for an Impressionist Effect. In SIGGRAPH ’97
Conference Proceedings.