IMAGE LIGHTING EFFECT MANIPULATION

FOR AN EFFICIENT STYLIZATION

Catherine Sauvaget and Vincent Boyer

L.I.A.S.D., Paris 8 University, 2 Rue de la Libert´e, 93526 Saint-Denis, France

Keywords:

Non-photorealistic Computer Graphics, Stylization, Lighting.

Abstract:

We propose a model to stylize images through the manipulation of different lighting effects and their styliza-

tion. Given an input image, we propose to generate semi-automatically a map containing each effect identiﬁed

by the user. Then, we propose different distortions and coloring to obtain artistic stylizations. Our model is

ﬂexible and well designed to help users who desire to stylize their images following light effects. This model

is intended for an image editing tool dedicated to comics stylizations.

1 INTRODUCTION

The purpose of images is to communicate a message

that can be emotional or informational. Artists use

lighting effects to enhance the desired information

and to create atmospheres bringing people to desired

psychological state.

Light has always been used in art to enhance spe-

ciﬁc parts of images. Light effects have been de-

scribed by J.M. Parram´on (Parram´on, 1987) and by

G.M. Roig (Roig, 2010). They depend on the light

source which transmission is direct or diffuse. Shad-

ows can be shadows of objects (shades) or projected

shadows representing shapes of objects (drop shad-

ows). The light can also produce dazzling effects due

to the reﬂective properties of the surface. As most

of viewers focus on colors used in a pictorial realiza-

tion, one can imagine that lighting effects are limited

to artistic movement techniques. Classical examples

are: Chiaroscuro consisting in violent contrasts be-

tween light and shadow to attract the viewer eye on

a speciﬁc part of the painting; Impressionism empha-

sizing the light: shadows are represented by saturated

colors and smooth diminution of light when illumi-

nated objects are in pastel colors; Comics using var-

ious styles to depict shadows: complementary colors

which are opposed on the chromatic hue wheel (It-

ten, 1961), hatching (Duc, 1983) and black ﬂat which

areas are very common in American comics (Mc-

Cloud, 1994). Dazzling effects are often represented

in comics as white areas with edges to enhance the

contrast (Duc, 1983). But for all these artistic move-

ments, the position, size, orientation, even more, pres-

ence or absence of each lighting effects such as drop

shadows are not obvious. Excepted for the hyper-

realistic movement, where lighting effects are repro-

duced with a high realism, every other pictorial move-

ments propose at most a plausible representation. As

an example, in ”The last Supper” by Leonardo da

Vinci, the lighting is plausible but physically unreal-

istic: where are the drop shadows? feet seem to be lit.

Even more, some comics vignettes contain strong ex-

aggerations. Artists sometimes willingly add shadows

to dramatize the image or on the contrary, to stylize

images, they remove some shadows to avoid an over-

loaded scene. Before considering any stylisation of

lighting effects, an artist has to decide their existence

and representation (shape, position...).

We provide a solution able to stylize lighting ef-

fects but also able to help the user to manipulate

them. Starting from a 2D input image, we pro-

duce semi-automatically a map containing different

kinds of lighting effects. Each effect can be shifted,

turned and distorted. Based on an artistic analysis

and this map, we propose deformations and six styl-

izations involving well-known artistic styles ranging

from chiaroscuro to comics. In the following, we

present previous work, and then our lighting effects

representation map. We detail our different stages:

distortion and stylization. Finally, our results are

given and commented.

299

Sauvaget C. and Boyer V..

IMAGE LIGHTING EFFECT MANIPULATION FOR AN EFFICIENT STYLIZATION.

DOI: 10.5220/0003822902990303

In Proceedings of the International Conference on Computer Graphics Theory and Applications (GRAPP-2012), pages 299-303

ISBN: 978-989-8565-02-0

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

2 PREVIOUS WORK

(Ibrahim and Anupama, 2005) and (Cavallaro et al.,

2004) detect shadow in videos or image sequences.

Based on retinex theory, (Sun et al., 2008) detect and

remove shadow from a single image. The work of

Ortiz (Ortiz, 2007) permits to remove dazzling ef-

fects from photographs. Unfortunately, these detec-

tions have a high computing cost and do not permit to

distinguish the different kinds of lighting effects pro-

duced by light.

Some research propose to distort images or spe-

ciﬁc objects in a scene. Carroll et al. (Carroll et al.,

2010) have proposed to change the perspective of an

image using vanishing points and lines that can be

modiﬁed by the user. This work is interesting but do

not permit to change the appearance of a speciﬁc ob-

ject in the image. Tobita (Tobita, 2010) has presented

a non-automatic solution dedicated to exaggerations

as the ones we can ﬁnd in mangas. Two main defor-

mations are introduced: on the background (blur . ..)

and on persons whose position and size can be modi-

ﬁed. They are mainly based on ﬁsh-eye deformations

bending sharply anything that does not pass through

the center of the circle.

Stylizations of lighting effects have been pro-

posed. Image-based methods have been proposed

to display soft shadows from 3D models (Agrawala

et al., 2000). Praun et al. (Praun et al., 2001) pro-

posed a system that creates hatching strokes in 3D

scenes. This kind of drawing conveys lighting and

properties of the material but we do not have the same

information in 2D. Only Sauvaget et al. (Sauvaget and

Boyer, 2010) have proposed an image-based styliza-

tion model based on a shadow map and in which six

stylizations can be combined. In this paper they do

not propose to shift or distort the lighting effects. We

use these stylizations on our lighting effect manipula-

tion model.

3 OUR MODEL

We present a new approach to manipulate and stylize

the different lighting effects of an image. Our model

permits to re-lit the previous shadow locations which

have been manipulated. We extend the model pro-

posed by Sauvaget et al. We present brieﬂy the model

to guide the reader and only detail hereafter our con-

tribution.

3.1 Map Creation

We represent lighting effects with a map which has

the size of the image and where each kind of lighting

effects is represented by a color. Note that we offer

the user to reﬁne it manually. The map is created in

two steps: detection, reﬁnement.

3.1.1 Detection

A shadow is a decrease of the light intensity. We use

the HLS model with L1 norm which is well designed

for shadow detection (Angulo, 2004). As depicted in

Sauvaget et al., we consider the the maximal and me-

dian values of lightness. We compute G the global

threshold with I the number of intensity levels(see

formula 1):

G = (max− med) ×

med

(1)

250

2550

med

max

med

max

255

maxmed

250200

100 150

100

150

58.8

39.2

19.6

Figure 1: Threshold detection examples.

With this method, only a part of pixels which

is smaller than the lightness threshold is considered.

The smaller the distance is between max and med

(in red ﬁgure 1), the smaller the considered area is

as well and avoid to take into account the highest

lightness values as shadow. This area is weighted by

the median value inﬂuencing directly the size of the

new area (in blue ﬁgure 1). The smaller med is, the

more shadow pixels are grouped and the more im-

portant reduction of threshold is needed. For now,

these pixels are considered as shadow without dis-

tinction of kind, the other pixels are considered as lit

parts in the image. Some results of this detection are

shown in section 4. As we propose to distinguish hard

and soft shadows, we choose to binarize the map for

hard shadows (black/white) or to preserve the original

lightness values of the original image for soft shad-

ows.

3.1.2 Reﬁnement

As in Sauvaget et al., to distinguish shades and drop

shadows, the user is invited to keep the previous black

color (or grey level shading) for the shades and to put

in blue the drop shadows. If some pixels are black

in the original image, our method does not permit to

distinguish them from shadows in the original image.

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

300

In such a case, the user can reﬁne the map adding

them in the lit parts (in white). To specify the dazzling

effects in the map, a red color is used.

3.2 Manipulation

Once the map is created, the user is allowed to modify

the place and the shape of lighting effects. With such

modiﬁcations, the image is changed and ”empty” ar-

eas may appear due to the displacement of an effect.

We present our displacement and deformation stage,

then we explain how we ﬁll these empty areas.

3.2.1 Displacement and Deformation

A bounding box is created during the selection of a

region deﬁning a lighting effect in the map. Its size is

max

− x

min

× y

max

− y

min

. Only the part of the image

corresponding to the selected lighting effect is kept in

the quad. The rest of the quad is transparent (see ﬁg-

ure 2). When placed on the image, the quad may un-

dergo classical transformations in three dimensions.

Moreover, artists sometimes remove shadows to sim-

plify the image. Our model allows this by removing

the selection.

The second image of ﬁgure 2 presents our map

with user reﬁnements for drop shadows. The bound-

ing box (in green) encompassing the drop shadowrep-

resents the selection. The previous location of the ef-

fect must be ﬁlled as if it was an lit part.

Note that deformations proposed ﬁgure 3 are ac-

tually possible with our system. Starting with the

left image, automatic transformations can provide the

center image. For the image on the right, the user

would have to reﬁne the map since the polygon should

be deformed.

Figure 2: Geopoliticus Child Watching the Birth of the

New Man by Salvador Dali (1934); map with the selected

shadow (quad in red); displacement and rotation on x.

Figure 3: Example of a drop shadow on a wall; possible

transformation; possible transformation with reﬁnement.

3.2.2 Space Filling

We propose two methods to transform the previous

location of the effect in a lit part of the image. The

ﬁrst method consists in a comparison between neigh-

bor surrounding pixels of the effect and the pixels of

the effect. We search for the neighbor surrounding

pixel which has the closest hue and saturation to the

mean hue and saturation of the pixels of the effect.

Once we ﬁnd the best pixel, we apply its lightness to

the selection shape. This lightness is applied every-

where in our selection implies the probability to cre-

ate ﬂat region : the volume previously existing in the

shadow is no more in this new lit part. This method is

well adapted in ﬂat colored textures but not if shading

exist. To preserve the volume, we propose a second

method consisting in calculating in the input image

the difference between lit parts and shadowparts from

our map. This difference is added to each pixel of our

selection. With this method, the volume of the shape

is preserved.

Remark that these methods cannot guaranty that

an area too desaturated or dark is rebuilt correctly be-

cause we only re-light and do not change the satura-

tion. Figure 4 illustrates this problem. On the left,

the original image with a yellow background HLS

(60, 90, 100). The second image corresponds to the

map and we select the top left shadow which is a

dark yellow. The last images show the result of each

method. With the ﬁrst method (left image), the closest

value is the yellow background color (L=90) and this

value is given to our selection. We obtain a color close

to a white HLS (60, 90, 18), because our selection has

a low saturation. With the second method, adding the

difference between lit parts and shadow parts gives

us the following color: HLS (60, 83, 18). None of

these methods provide a suitable solution. The ex-

treme case is black color HLS (0, 0, 0), where the hue

is red deﬁned.

Figure 4: Original; map; ﬁrst method result; second method

result.

3.2.3 Style Propositions

We propose six different stylizations following

Sauvaget et al. (Sauvaget and Boyer, 2010):

chiaroscuro, impressionism, complementary color,

hatching, black ﬂat areas and dazzling effects. The

IMAGE LIGHTING EFFECT MANIPULATION FOR AN EFFICIENT STYLIZATION

301

user chooses to apply the stylization on a speciﬁc

component (effect) of the map.

4 RESULTS

We present some results obtained with our model. All

of these images have been produced in real-time on a

Pentium 2.5GHz with 3Go of memory.

Figure 5 presents an original image from Le Pixx,

the map and the lighting effect transformation results

on the drop shadow of the chair. The result of the clos-

est method to ﬁll in the previous location is shown.

Figure 6 shows a mixed result from our lighting ef-

fect stylization model and from comics stylization

model (Sauvaget and Boyer, 2008).

An assessment protocol has been realized on our

results. Ten persons (novices, experts in computer

graphics and illustrators) evaluated ﬁfty images (in-

terior and exterior scenes, illustrations...).

We would like to know if the users succeed in

creating the stylization they desired with the existing

possibilities of our tool. All of them found intuitive to

shift, rotate and distort the lighting effects. However,

when not combining some of our lighting styliza-

tions with the comics stylization model of Sauvaget

et al. that permits to obtain a global coherence in the

stylization of the image, 90% of them felt disturbed.

Artists felt limited by the actual number of possible

shadow stylizations but they appreciated the mix be-

tween the atmosphere and light effects (see ﬁgure 6).

Figure 5: Original; map; result.

Figure 6: Original; map; result mixing our model and

Sauvaget et al. comics stylization one.

5 CONCLUSIONS

We have proposed a model to manipulate and stylize

lighting effects for 2D images. The principal limit of

the detection method is that dark objects are detected

as shadow. Our model permits a visual and semantic

distinction between the lighting effects. It is ﬂexible

and allows different stylizations on different lighting

effects.

In future work, we will improve our model by

adding more stylizations and colored light effects. We

plan to add existing effects like hatching using gradi-

ents. We also plan to consider coupling our approach

with the depth map produced by (Sauvaget and Boyer,

2008) to enhance the contrast between the different

kinds of lighting.

REFERENCES

Agrawala, M., Ramamoorthi, R., Heirich, A., and Moll, L.

(2000). Efﬁcient image-based methods for rendering

soft shadows.

Angulo, J.; Serra, J. (2004). Traitements des images de

couleur en repr´esentation luminance/saturation/teinte

par norme l1. In Traitement du signal, pages 583–604.

Carroll, R., Agarwala, A., and Agrawala, M. (2010). Im-

age warps for artistic perspective manipulation. ACM

Trans. Graph., 29:127:1–127:9.

Cavallaro, A., Salvador, E., and Ebrahimi, T. (2004). De-

tecting shadows in images sequences. In European

Conference on Visual Media Production, pages 165 –

174.

Duc, B. (1983). L’art de la BD. Gl´enat et B. Duc, gl´enat et

b. duc edition.

Ibrahim, M. and Anupama, R. (2005). Scene adaptive

shadow detection algorithm. In Proceedings Of World

Academy Of Science, Engineering and Technology,

pages 1307–6884.

Itten, J. (1961). Kunst der Farbe. Ravensburg: Otto Maier

Verlag.

McCloud, S. (1994). Understanding Comics the invisible

art. Harper Paperbacks.

Ortiz, F. (2007). Real-time elimination of brightness in

color images by ms diagram and mathematical mor-

phology. In Proceedings of the 12th international

conference on Computer analysis of images and pat-

terns, CAIP’07, pages 458–465, Berlin, Heidelberg.

Springer-Verlag.

Parram´on, J. (1987). Ombres et lumi`eres dans le dessin et

la peinture. Bordas.

Praun, E., Hoppe, H., Webb, M., and Finkelstein, A. (2001).

Real-time hatching. In Proceedings of the 28th an-

nual conference on Computer graphics and interac-

tive techniques, SIGGRAPH ’01, pages 581–, New

York, NY, USA. ACM.

Roig, G. (2010). Peindre la lumi`ere. Guides Oskar.

Sauvaget, C. and Boyer, V. (2008). Comics stylization from

photographs. In Proceedings of the 4th International

Symposium on Advances in Visual Computing, ISVC

’08, pages 1125–1134, Berlin, Heidelberg. Springer-

Verlag.

Sauvaget, C. and Boyer, V. (2010). Stylization of light-

ing effects for images. Signal-Image Technologies

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

302

and Internet-Based System, International IEEE Con-

ference on, 0:43–50.

Sun, J., Du, Y., and Tang, Y. (2008). Shadow detection

and removal from solo natural image based on retinex

theory, volume 5314 of Lecture Notes in Computer

Science. Springer-Verlag, Berlin Heidelberg.

Tobita, H. (2010). Enformanga: interactive comic creation

with deformation. In Proceedings of the International

Conference on Advanced Visual Interfaces, AVI ’10,

pages 397–398, New York, NY, USA. ACM.

IMAGE LIGHTING EFFECT MANIPULATION FOR AN EFFICIENT STYLIZATION

303