AUTOMATIC COMPUTER GENERATION OF STIPPLING

ILLUSTRATIONS WITH FELT-TIP COLOURED PENS

Germ´an Arroyo and Domingo Mart´ın

Dep. Lenguajes y Sistemas Inform´aticos, University of Granada

C/ Periodista Daniel Saucedo Aranda s/n, 18071, Granada, Spain

Keywords:

Rendering, Non-photorealistic rendering, Painting-like rendering, Drawing.

Abstract:

Nowadays, non-photorealistic rendering is an area that not only focuses on simulates what artist do or the tools

they use, but also on generates new expressive tools for digital art. In this paper we present a new algorithm

to generate beautiful stippling illustrations with felt-tipped colour pen from a photograph or an image. This

technique is not actually used by artists because technical limitations, therefore this new algorithm might be

helpful. We introduce a novel stochastic approach to place coloured dots in a speciﬁc order based on the

information of the contrast, borders and histogram of the input image. The system is able to generate an

unlimited number of non regular synthetic colour dots without the necessity of being scanned. These dots will

be composed in a speciﬁc order to generate the ﬁnal illustration.

1 INTRODUCTION

Stippling is the technique of drawing using dots,

which are composed of pigment in a single colour ap-

plied with a pen or a brush, changing the density to

obtain different shades. The stippling technique can

be altered to use colours. This technique must not be

confused with pointillism, which uses small distinct

dots of colour to create the impression of a wide se-

lection of other colours and blending. The technique

of coloured stippling overlaps the dots to shade the

illustration whereas it is not allowed in pointillism.

Several problems happen when an artist try to stip-

ple an illustration using felt-tip pens. The ﬁrst prob-

lem is that the number of different colours of these

pens are very limitted. A second problem is that the

amount of ink and the porosity of the tip makes expen-

sive to stipple a complete illustration, specially with

medium-tip markers. The paper is also a problem be-

cause a thin paper cannot admit a large amount of ink,

and it brokes; on the other hand, a thick paper spread

out the ink too much, in such a way that the shapes

become blurred and undeﬁned.

This paper is structured as follows: In Section 2

we discuss related work. Section 3 we present an

overview of our system. In Section 4 we explain the

algorithms in detail. Section 5 discusses the results of

our approach. The paper is concluded in Section 6.

2 PREVIOUS WORKS

Abstract representation of still images was intro-

duced by Haeberli(Haeberly, 1990) using image color

gradient and user interactivity for painting. Hertz-

mann(Hertzmann, 1998) places curved brush strokes

of multiple sizes on images for painterly rendering.

The technique ﬁlls color by using big strokes in the

middle of a region and uses progressively smaller

strokes as one approaches the edges of the region.

Shiraishi and Yamaguchi(Shiraishi and Yamaguchi,

2000) improves the performance of above method

by approximating the continuous strokes by place-

ment of rectangular strokes discreetly along the edges

to create painterly appearance. Santella and De-

Carlo(Santella and DeCarlo, 2002) uses eye track-

ing data to get points of focus on images and cre-

ate painterly rendering with focus information. There

good works for painting terrains(Coconu et al., 2006;

Bhattacharjee and Narayanan, 2008), but they do

not work with general models. Rudolf el al. pro-

poses a physically-inspired model to simulate wax

crayons(Rudolf et al., 2003). All these techniques

work well on single images but do not simulate colour

263

Arroyo G. and Martín D. (2010).

AUTOMATIC COMPUTER GENERATION OF STIPPLING ILLUSTRATIONS WITH FELT-TIP COLOURED PENS.

In Proceedings of the International Conference on Computer Graphics Theory and Applications, pages 263-266

DOI: 10.5220/0002821002630266

 SciTePress

stippling.

Most of the related research on stippling is fo-

cused on generating dots according to the shading

of a photograph, paying almost no attention to the

shape of dots or the techniques that artists use(Secord

et al., 2002). Some methods propose using cir-

cles instead of realistic dots(Mould, 2007)(Gooch and

Gooch, 2001)(Schlechtweg et al., 2005)(Yuan et al.,

)(Lu et al., 2003). This differs noticeably from the

illustrations created by artists because natural dots

have a gradient. Other methods focus on distribut-

ing the dots correctly along a surface according to the

shading(Secord, 2002)(Pastor et al., 2004). In these

cases, the use of Central Voronoy diagrams produce

easily recognisable patterns. Renderbots is based on

the idea of particles but the results, when they simu-

late stippling, have the same problems with patterns

too(Schlechtweg et al., 2005). None of them uses

coloured dots because it is a less usual technique.

In the next Section, we will expose our solution to

stipple with simulated felt-tip pens.

3 OVERVIEW

First, we will discuss the proposed system. It

is based on the scheme presented in Figure 1.

As we can see, the algorithm has the following

steps:

1: The information is obtained from the input image.

2: A matrix with probability is generated from this

information.

3: The algorithm enters in a loop:

4: loop

5: The new place and size of the dot is computed.

6: A new dot is generated with a simpliﬁed colour

of the region.

7: The dot is composed in the output image.

8: end loop

This algorithm ﬁnishes when the stop condition is

reached or the user stops it. A matrix of probability

will guide all the process to place the dots, in such a

way that the order of the dots placement is determined

by it.

4 ALGORITHMS

The algorithm can be divided in several subalgorithms

that takes an input and returns an output to the next

step. These algorithms are explained in the following

subsections.

Figure 1: An overview of the algorithm.

4.1 Obtaining Information

First of all, it is necessary to ﬁlter the photograph

to obtain relevant information before using the algo-

rithm to place dots. This ﬁltering process is based

on what artists do before they draw a stipple illustra-

tion. The process may vary from artist to artist, but

in general, they begin marking silhouettes and stip-

pling them. The next step is to stipple on the darkest

areas of the image. It is also important to stipple in

areas where details can be enhanced and where there

are different or relevant elements of the photograph.

The background is usually irrelevant, so, the most re-

peated tone in the photograph is ignored by the artist.

Colour stippling is not an exception to this rule.

4.2 Generating the Function of

Probability

A discrete representation of a probability density

function (PDF) is constructed with the information

obtained in the previous step. The user adjusts the

relevance of the histogram, the contrasted image and

the borders in the PDF by hand. Hence, a 2d array

with the values of the PDF is initialized.

Once the array has been generated, the next step

is iterate until all the dots have placed.

GRAPP 2010 - International Conference on Computer Graphics Theory and Applications

264

4.3 Algorithm for Positioning Dots

The algorithm for dots placement removes the pat-

terns in the ﬁnal image. We use a method that is also

based on a stchocastic algorithm but including some

parameters that are controlled by the user. These pa-

rameters handle the number of placed dots and the

size of the dots.

The algorithm places the dot randomly according

to the valuesof the array 2d, in such a waythat highest

values has more probability of being chosed to place

a dot.

The dots can be placed in the output image using

the following formula for every pixel of the dot:

= (1− A) ·C

+ (A) ·C

(1)

Where C

is the colour destiny, C

is the colour

source, and A is the value of the intensity of the dot.

4.4 Algorithm for Generating Colour

Dots

The automatic generation of dots removes the artiﬁ-

cial appearance of the illustration. Before generating

a dot, the algorithm must decide what the colour will

be for all the colour. We must take in consideration

that each individual dot cannot contain more than one

color.

The colour is determined by a simple equation:

dot

pixel

· 3.0

pixel

.red +C

pixel

.green+C

pixel

.blue)

(2)

This normalization of the colour produces a ﬂat-

tening and a reduction of the brightness, which is used

in computer vision algorithms to simplify the illumi-

nation of the scene(Finlayson et al., 1998). Addition-

aly if the colour is almost black, the luminosity can

be increased, specially if the intention is to print the

image. The gray colours are substituted by only two

tones of grey. The reason is that grey colours do not

contribute with too much detail to the scenery and can

be obtained easily by stippling on the same place re-

peatedly.

The problem is reduced to generate a grey scaled

dot, and then multiply the obtained colour by the lev-

els of intensity of the dot. White values are not a prob-

lem because if a white colour is found, the algorithm

does not stipple there.

The proposed algorithm to obtain grey dots is

based on Monte Carlo methods. In these kinds of

methods the algorithms decide at every step when the

solution is false, but do not know when the solution is

true. Therefore, the algorithm iterates a certain num-

ber of steps until it reaches an approximate solution.

The algorithm is as follows:

1: Create a matrix (P) of local probabilities

2: Create a matrix (I) of intensity values

3: Create a matrix (A) of absorption values

4: Generate a seed

5: for i = 0 to i < N DROPS do

6: F

(n) returns a position (p

, p

)

7: Deposit the ink in the returned position

8: end for

9: Apply a Gaussian ﬁlter to smooth the result

N DROPS is the number of iterations of the algo-

rithm, it can be estimated from the desired size of the

dot. The size of the matrices P, I and A is the double

of the size of the dot. The matrix P is a probability

density function (PDF). I is the ﬁnal image of inten-

sities, whereas A is the abosorption of the paper, and

it can be obtained from a gradient image.

The seed can be simply a small circle, this circle

writes the matrix I at its center with a dark value. The

matrix P satisﬁes the discrete condition

∑

i, j

P(i, j) =

1 because it is a PDF. Therefore, the values of the

matrix P are initialized to 0 but in the places where

is the seed. These cells are initialized to an uniform

value scaled according to the number of used cells.

The ﬁnal dot is used in the algorithm described at

previous Section.

5 RESULTS

The algorithm produces a stippling illustration with

felt-tip pens that degrade the colours in an aesthetic

way. The algorithm also can detect how many dots

place in the ﬁnal illustration by using the information

of the contrasted image. If the background of the im-

age is more or less plain, it is detected and removed

from the photograph. The algorithm can also take as

input a rendered image from a 3D scene or painted il-

lustrations, it also works with complex photographies,

as shown in Figure2. It can be appreciated how the al-

gorithm produce good results even with very compex

images or photograph because the shape and colour

of the dots.

All these images has been generated with felt-tip

pens between 40mm and 60mm.

6 CONCLUSIONS AND FUTURE

WORKS

We have presented an algorithm to automatically

draw coloured illustrations and simulates felt-tip stip-

pling. We have developed a fast and direct equilibra-

AUTOMATIC COMPUTER GENERATION OF STIPPLING ILLUSTRATIONS WITH FELT-TIP COLOURED PENS

265

Figure 2: A illustration generated by our algorithm.

tion technique that is based on a probabilistic model.

This algorithm draws the dots in a natural order, just

like artists do. We have introduced an automatic con-

trol to stop the algorithm detecting when the illustra-

tion is ﬁnished. Our system provides both interactiv-

ity and high-quality output. It can take as input both

photographs, 3D rendered images and illustrations.

Future research is proposed into how more artistic

knowledge can be included automatically within the

application. A study of how to print this kind of im-

ages should be also interesting. Another important is-

sue of future work is the introduction of temporal co-

herence when applied to frames in a video, specially

when try to use the same colours when objects which

are moving in a scene.

ACKNOWLEDGEMENTS

We thank the Ministerio de Educaci´on y Ciencia of

Spain for funding part of this work under the project

TIN2007-67474-C03-02.

REFERENCES

Bhattacharjee, S. and Narayanan, P. J. (2008). Real-time

painterly rendering of terrains. In Proceedings of Sixth

Indian Conference on Computer Vision, Graphics &

Image Processing.

Coconu, L., Deussen, O., and Hege, H. C. (2006). Real-

time pen-and-ink illustration of landscapes. In NPAR

06: Proceedings of the 2nd symposium on Non-

photorealistic animation and rendering, pages 27–36.

Finlayson, G. D., Schiele, B., and Crowley, J. L. (1998).

Comprehensive colour image normalization. In ECCV

’98 : European conference on computer vision, vol-

ume 1407, pages 475–490.

Gooch, B. and Gooch, A. (2001). Non-photorealistic Ren-

dering. A. K. Peters.

Haeberly, P. (1990). Paint by numbers: abstract image rep-

resentations. In SIGGRAPH’90: Proceedings of the

17th annual conference on Computer graphics and in-

teractive techniques, pages 207–214.

Hertzmann, A. (1998). Painterly rendering with curved

brush strokes of multiple sizes. In Computer Graphics

(Annual Conferences), number 32, pages 453–460.

Lu, A., Morris, C., Taylor, J., Ebert, D., Rheingans, P.,

Hansen, C., and Hartner, M. (2003). Illustrative inter-

active stipple rendering. In IEEE Transactions on Vi-

sualization and Computer Graphics, volume 9, pages

127–138.

Mould, D. (2007). Stipple placement using distance in a

weighted graph. In Proceedings of Computational

Aesthetics in Graphics, number 3, page unknown.

Pastor, O. M., Freudenberg, B., and Strohthotte, T. (2004).

Real-time animated stippling. In Proceedings of

NPAR 2004, volume 23, pages 62–68.

Rudolf, D., Mould, D., and Neufeld, E. (2003). Simulat-

ing wax crayons. In Proceedings of the 11th Paciﬁc

Conference on Computer Graphics and Applications,

page 163.

Santella, A. and DeCarlo, D. (2002). Abstracted painterly

renderings using eye-tracking data. In NPAR 02: Pro-

ceedings of the 2nd symposium on Non-photorealistic

animation and rendering, pages 53–58.

Schlechtweg, S., Germer, T., and Strothotte, T. (2005). Ren-

derbots: Multi agent systems for direct image genera-

tion. Computer Graphics Forum, 24:283–290.

Secord, A. (2002). Weighted voronoi stippling. In Proceed-

ings of NPAR, pages 37–43. ACM Press.

Secord, A., Heidrich, W., and Streit, L. (2002). Fast prim-

itive distribution for illustration. In Thirteenth Euro-

graphics Workshop on Rendering, pages 215–226.

Shiraishi, M. and Yamaguchi, Y. (2000). An algo-

rithm for automatic painterly rendering based on lo-

cal source image approximation. In NPAR 00: Pro-

ceedings of the 1st international symposium on Non-

photorealistic animation and rendering, pages 53–58.

Yuan, X., Nguyen, M. X., Zhang, N., and Chen, B. Stip-

pling and silhouettes rendering in geometry-image

space. In Proceedings of Eurographics Symposium on

Rendering, pages 193–200.

GRAPP 2010 - International Conference on Computer Graphics Theory and Applications

266