SKIN MODELING AND RENDERING BASED ON VISUAL
PERCEPTION
Azam Bastanfard, Nadia Magnenat Thalmann
MIRALab, CUI University of Geneva,24 rue du General Dufour,CH-1211 Geneva SWITZERLAND
Keywords: Skin rendering, Human perception, Modelling.
Abstract: Human skin modelling and rendering are affected with a variety of cues. These are including human visual
perception of skin texture and lighting. An attempt to mimic such attributes by computer is an aspiring goal
and challenging task. This paper proposes a novel algorithm with two techniques as a key solution capturing
such a variety of cues to skin appearance. The idea is to capture these two characteristics for skin rendering.
The first is the texture generation that developed in visual perception. The second is skin texture rendering.
These techniques discuss the skin noise simulation based on human perception theory and simulate skin
noise texture. Then, the skin is rendered with what we call the Bidirectional Reflectance Distribution
Function Texture Magnitude technique. The original contribution and advantages of this paper compared
with other proposed methods are simple to implement, reliable and their computations are fast enough for an
interactive environment. Experimental results demonstrate our approach for skin texture generation.
1 INTRODUCTION
Creating realistic rendering of human skin has an
endeavour in computer graphics for nearly three
decades (Nahas et al., 1990). Application of human
skin modelling and rendering are founded in
sophisticated computer interface, VR applications,
computer games, multimedia and in a broad variety
of production animation. Human visual system can
easily recognize skin with other materials.
Reproducing visual perception by computer is a
difficult task because of the complex and individual
texture of the skin and spatially varying reflectance
properties of skin.
Rendering human skin based on human visual
perception requires a handle on how to capture the
image space created by the source of variation. In
principle two main elements have to be considered
in skin texture rendering based on human vision and
skin reflectance. This paper shows how to simulate
skin accounting for these elements. Our approach is
motivated by a research on two studies.
The first is human visual perception and noise
importance and texture generation for simulating
skin texture. The second is the rendering skin texture
based on a new BRDF technique.
The underlying idea to know the noise
importance and its limit is useful to calculate ideal
performance as a benchmark for simulation and
rendering. Visual sensitivity is a product of two
factors that isolate visual process more easily (Malik
et al., 1990). These two factors contain observer
efficiency and equivalent noise. On the other hand
the ability to delineate characters skin in visual
world depends partly upon the perception of noise
consistency.
Perlin has determined a function, which uses an
interpolation to combine noise-generating functions
(Perlin, 1985). It has also been used to produce
natural textures. Finally we render skin texture based
on BRDF technique. In rendering phase generated
texture is applied to render skin that will be
discussed in section 4.2.
This paper organized as the followings. Section
2 reviews some of the prior researches in skin
texture modelling and rendering with some
discussions on its problem. Section 3 provides a
brief summary of human vision and skin texture
perception. Section 4, introduces our proposed
techniques for skin texture generation and its
reflectance. Some experimental results and
discussion are given in section 5. Finally, conclusion
and direction for future work are discussed in
section 6.
313
Bastanfard A. and Magnenat Thalmann N. (2006).
SKIN MODELING AND RENDERING BASED ON VISUAL PERCEPTION.
In Proceedings of the First International Conference on Computer Graphics Theory and Applications, pages 313-318
DOI: 10.5220/0001357503130318
Copyright
c
SciTePress
2 PRIOR WORKS AND
PROBLEMS
Human skin modelling and rendering are extremely
important in enhancing the realism of human
appearance. Varied models are used to investigation
on human skin and researchers discussed it from
different point of view (Freefberg,
1999)(Hiraoka,1993). For example there are
geometric models, physically based models and
biomechanical models using either particle system
or continues system. Ishii et al. (Ishii, 1993) also
generated human skin texture, based on optical
scattering in the skin layer with a 3D mesh structure
of the skin surface. They reproduced the change in
texture appearance for various scattering coefficients
in the skin.
Debevec et al. (Debevec, 2000) proposed a
technique to obtain the reflectance field of a face at
each point by rotating the light source around the
face. Hanrahan and Krueger (Hanrahan, 1993) have
modelled skin subsurface structure and photon
migration in the skin, and have simulated the
shading of skin colour. Jensen et al.(Jensen, 2001)
reproduced realistic facial images based on photon
mapping or the diffusion equation technique for
subsurface scattering.
Bastanfard et al (Bastanfard, 2004) proposed a
technique for skin cosmetics based on wrinkle
removal using inpainting.
Cula et al. (Cula, 2005) developed a texture
representation technique that counts for changes of
skin as imaging condition are varied. An E-cosmetic
function for digital images based on physics and
physiologically-based image processing was
proposed by Tsuruma et al. (Tsuruma, 2004). Other
extension and improvement have been reported in
(Marschner, 1997)(Ramamoorthi, 2001)(Sato,
1994).
A Dynamic wrinkle model and skin aging was
proposed by Wu and Thalmann (Wu, 1995). Another
approach has been proposed by Boissieux et al.
(Boissieux, 2000) to simulate skin aging and
wrinkles with cosmetics insight. In this method
texture mapping offers a good simulation of static
wrinkles, but constructing visually interesting bump
maps requires practice and artistic skills.
Although much work has been done in skin
rendering, perception based model on description of
skin does not exist because of the complexity. On
the other hand there are no methods in which the
texture image generates, then modified as a function
for skin rendering. In this paper, a texture
analysis/synthesis technique is used to change the
amount of texture spatially then a Bidirectional
reflectance distribution function texture magnitude is
proposed to rendering effect. The results
demonstrate realistic modelling and rendering of
skin texture.
The original contribution of this paper compared
with the previous mentioned work is given in terms
of following advantages. First the proposed
techniques are efficient in time complexity and
simple to implement. Second they are local. By local
we mean that we can generate different skin based
on local noise. Third they provide a simple way to
capture the skin surface details. Finally, the
computations are fast enough for an interactive
environment.
3 VISUAL PERCEPTION OF
TEXTURE
Modelling of human skin has been known as an
essential step of different problems. Based on this
need many attempts have been done in this regard.
On the other hand, the ability to delicate skin texture
in computer graphics and virtual worlds depends
partly upon the perception of textural consistency.
The theory of texture perception attribute pre-
attentive texture discrimination to differences in
first-order statistics of stimulus features such as
orientation, size and brightness of constituent
elements.
The reason to describe particular aspect of visual
perception like contrast is human vision sensitivity
factor due to achieve realism in computer graphics.
For instance a local contrast calculation produces
contrast image that describe the response of
ensemble of neuron to a particular spatial frequency
and orientation. Contrast energy E is the square of
the contrast function summed over the dimensions
along which the stimulus varies (Pelli, 1999). For
static two-dimensional stimuli, signal energy is
generated over space:
dxdyyxcE ),(
2
∫
∫
=
(1)
The contrast function is normalized derivation of the
luminance function from the background level,
[
]
bb
LLyxLyxc /),(),(
−
=
(2)
We call them the channel images as they
represent different channel of visual system.
GRAPP 2006 - COMPUTER GRAPHICS THEORY AND APPLICATIONS
314
Therefore visual sensitivity can be proportioned into
observer efficiency and visual noise (Pelli, 1999). In
this paper we generate noise texture and new
technique on texture luminance to gain geometry of
skin texture.
4 OUR PROPOSED ALGORITHM
The quest for improving realism is always an
important goal of computer graphic and virtual
world. Therefore current models of skin appearance
are not sufficient to support the demands for fast
algorithms. One major difficulty in skin perceptual
appearance is the extreme complexity of real skin
that exhibit many subtle variations over its entire
surface. This section explores a new algorithm with
two techniques for modelling and rendering of skin
based on human perception. In this approach Perlin
method is used to generate a 2D skin texture. Then
we extract noise from texture by proposing BRDF
texture magnitude. Finally skin texture will be
generated. Figure 1 illustrates the steps of our
proposed algorithm.
Figure 1: General structure algorithm outline.
4.1 Skin Rendering
Over the last few years, many construction of BRDF
have been introduced in optics and in computer
graphics literature (Kautz, 2004)( Kobbelt,
1997)(Marschner, 1999). This section focuses on
rendering of skin texture. Our idea is derived from
the geometry of bidirectional reflectance diffusion
function BRDF and the concept of texture analysis/
synthesis in term of textures magnitude. In the
followings we will summarize the fundamental
definition and basic notation of BRDF, texture
magnitude function and finally introduce our
proposed technique.
4.2 Bidirectional Reflection
Distribution Function
The bidirectional reflectance distribution function
(BRDF) of a surface describes how light is scattered
at its surface. Its value measures the ratio of the
radiance L exiting the surface in a given direction to
the incident irradiance I a particular wavelength
λ
from an incident solid angle
i
d
ω
about a given
illumination direction. The BRDF denoted by
r
f is
according to figure 2, as the following;
()
iiiii
eer
eeiir
dL
L
f
ωθφθ
φ
θ
λφθφθ
cos),(
),(
,,;, =
(3)
Where,
),(),(
,, eevreesrr
LLL
φ
θ
φ
θ
+
=
(4)
Is the reflected radiance which has two
components, one is the reflected radiance due to the
surface scattering
sr
L
,
, and the other
component
vr
L
,
is due to subsurface volume
scattering. Therefore it has five variables but its
domain is reduced somewhat by a symmetry called
reciprocity. Reciprocity stated that reversing the
light path doesn’t change the reflectance, that is
(
)
),,,,(,,;,
11222211
λ
φ
θ
φ
θ
λ
φ
θ
φ
θ
rr
ff
=
(5)
Figure 2: Geometry of surface reflection.
The units of BRDF are thus inverse steradians. It
has been widely adopted where
ii
φ
θ
, are the
SKIN MODELING AND RENDERING BASED ON VISUAL PERCEPTION
315
incident zenith and azimuth angles and
ee
φ
θ
, are
the corresponding reflection angles.
Intuitively the BRDF represents for each
incoming angle, the amount of light that is scattered
in each outgoing angle. For a Lambertian (perfectly
diffuse) surface, for example, the BRDF is constant,
and equal to
π
ρ
where
ρ
is the diffuse reflection
coefficient and the factor of
π
is necessary so that
the BRDF is correctly normalized. From this point
of view then for a diffused surface, skin texture can
be rendered by texture magnitude that will explain in
next section.
4.3 BRDF Texture Magnitude
Skin data collection is time consuming and tedious
task. Therefore we generate texture using Perlin
noise.
In this approach we do not need to have a lot of
skin sample. The high spatial sampling of the
generated texture used to determine the BRDF
inherently enables extraction of the BRDFTM.
Figure 3 illustrates 2D skin texture generated using
Perlin noise. For example this 500 × 500 pixel
textures has an average digital count for the R, G
and B channels of 252, 215 and 190, this is
proportional to the BRDF.
Figure 3: 2D skin generated texture.
We implement BRDF for texture magnitude be
computed by convolution of the BRDF texture.
∑∑
−
=
−
=
−−=
1
0
1
0
2
],[],[
1
],[
m
i
n
j
jyixhyxf
x
yxg
(6)
Where f[x,y] is the texture and h[x,y] is a
convolution kernel. The g[x,y] histograms use
BRDFs texture data.
Therefore we made a new fast and empirical
algorithm for skin texture rendering.
In this approach at first we make a texture using
Perlin method. Since the texture colour ratio will be
changed, we called it texture magnitude. Then we
convert texture magnitude to BRDF average. We
have colour dispersion when the RGB refraction
indices are different. The difference in colour
between the images correspond changes gives the
texture magnitude, which can create complex
shading.
The process of our algorithm yields skin images
with fine appearance in a simple way. In this method
we control the skin noise and skin surface, surface
variation by adjust the noise range in texture.
Therefore, by generating Perlin noise texture and
proposing BRDFTM technique we generated skin
surface based on visual perception. This modelling
and rendering of human skin is fast and it is
convenient for interactive environments.
5 EXPERIMENTAL RESULTS
In this section, we present some images from
implementation of our proposed technique. Figure 4
demonstrates a virtual human with our skin texture
generation algorithm.
We illustrate generated skin texture with near
view in figure 5. The skin texture parameter contains
252,215,190 for R, G, and B. The period of Perlin
noise in this image is 1.56 and limit of noise is 10.
Figure 6. illustrates skin texture on more details.
In this approach at first we make a texture,
change the colour ratio that we called it texture
magnitude. Then we convert texture magnitude to
BRDF average. Therefore by generating different
skin noise texture, skin surface will be generated.
How ever our approach is not the exact
rendering of human skin but it is fast and it is
convenient for interactive environments. Figures
show skin texture with more convincing appearance
by proposed method.
Figure 4: Proposed skin rendering.
GRAPP 2006 - COMPUTER GRAPHICS THEORY AND APPLICATIONS
316
Figure 5: Skin texture.
Figure 6: Skin texture in more details.
6 CONCLUSION AND FUTURE
WORK
Modelling and rendering of human skin texture are
difficult task and challenging problem. Many
different approaches have been proposed for human
skin modelling and rendering. This paper presents a
new and effective skin texture rendering. Two
proposed techniques have been introduced one for
empirical texture modelling and other for the
rendering effects. The advantages of the proposed
technique over previous method are given as
follows. First, the proposed techniques are efficient
in time complexity, simple to implement, and
reliable in which they don’t need to collect lot of
data. Second they provide a simple way to capture
the geometrical details founded in generated texture
without any constraints and render them smoothly.
Finally their computations are fast enough for an
interactive environment. On the other hand the
definition of skin texture appearance is a difficult
task. It depends not only on the structure of skin but
also so many aspects of one’s life, including
climatic, Psychology, and other parameters.
Therefore for more realistic skin effect further
research effort is needed. We would like to extend
our algorithm based on spherical wavelets.
ACKNOWLEDGMENTS
Special thanks to Nedjma Cadi Yazli for preparing
the model.
REFERENCES
Bastanfard A., Bastanfard O., Takahashi H., M.
Nakajima M., 2004. Toward anthropometrics
simulation of face rejuvenation and skin cosmetic. In
Journal of Visualization and Computer Animation.
15(3-4): 347-352.
Boissieux, L., Kiss, G., Thalman, N., and Kalra, P., 2000.
Simulation of skin aging and wrinkles with cosmetics
insight”, In Eurographics Workshop on Animation and
Simulation (EGCAS 2000), pp. 15– 28.
Cula O.G, Dana K. J., Murphy F. P., Rao B. K., 2005.
Skin Texture Modelling. In International Journal of
Computer Vision 62(1-2): 97-119.
Debevec P., Hawkins T., Tchou C., Duiker H.-P., Sarokin
W., Sagar M., 2000. Acquiring the reflectance field of
a human face. In Proceedings of SIGGRAPH 2000,
145-156.
Freefberg I.M.,Eisen A.Z., Wolfe K., Austen F. K.,
Goldsmith L.A., Katz S.I. Fitspatrick T.B.,1999.
Fitzpatrick’s Dermatology in General Medicine. The
McGraw Hill.
Hanrahan P., Krueger W., 93. Reflection from layered
surfaces due to subsurface scattering. In Proceedings
of ACM SIGGRAPH '93, ACM Press / Computer
Graphics Proceedings, Annual Conference Series,
ACM, 187-194.
Hiraoka M., Firbank M., Essenpreis M., Cope M., Arrige
S. R., Zee P. V. D., Delpy D. T., 1993. A Monte
Carlo investigation of optical path length in
inhomogeneous tissue and its application to near-
infrared spectroscopy. In Phys. Med. Biol. 38, 1859-
1876.
Ishii T., Yasuda T., Yokoi S., Toriwaki J., 1993. A
generation model for human skin texture. In
Proceedings of CG International 1993, Springer-
Verlag, 139-150.
Jensen W., Maeshner S. R., Levoy M., AND Hanrahan P.
,2001, A practical model of subsurface light transport.
In Proceedings of ACM SIGGRAPH 2001, 551-518.
Kautz J., Sarlette R., Klein H., Sidel H ., 2004.
Decoupling BRDF s from Surface Mesostructures,
Graphic interface 2004, May 2004, 177-184.
Kobbelt
L., Stamminger M., Seidel H.-P., 1997, Using
subdivision on hierarchical data to reconstruct
radiosity distribution. In Computer Graphics Forum ,
Proc. Eurographics’97,
16, 3 (1997), C347–C355.
Marschner S. R., Greenberg D. P., 1997. Inverse lighting
for photography. In Proceeding of Fifth IS&T/SID
Color Imaging Conference1997, IS&T and SID, 262-
265.
Marschner S. R., Westin S. H., Lafortune E. P. F.,
Torrance K. E., Greenburg D. P.,1999. Image based
BRDF measurement including human skin. In
Proceedings of 10
th
Eurographics Workshop on
Rendering, 1999, 139-152.
Nahas, M., Huitric, H., Rioux, M., and Domey, J. 1990.
Facial image synthesis using skin texture recording.
Visual Computer, 6(6):337– 343.
SKIN MODELING AND RENDERING BASED ON VISUAL PERCEPTION
317
Perlin K., 1985. An image synthesizer. In Proceedings of
ACM Siggraph 1985, 19(3)287-296.
Ramamoorthi R., Hanrahan P.,2001. A signal processing
framework for inverse rendering. In Proceedings of
ACM SIGGRAPH 2001, 117-128.
Sato Y. Ikeuchi K. Temporal-color space analysis of
reflection, 1994. In Journal of Optical Society of
America A1994, 11, 11, 2990-3002.
Pelli, D.G, Farell, B., 1999. Why use noise?, In Journal of
optical Society of America A, 16.3, 647-653.
Tsumura N., Ojima N., Sato K., Shiraishi M., Shimizu H.,
Nabeshima H., Akazaki S., Hori K.,Miyake Y., 2004.
Image-based skin color and texture analysis/synthesis
by extracting hemoglobin and melanin information in
the skin. In Proceedings of ACM Siggraph 2004,
19(3)770-779.
Wu Y. and Magnenat Thalmann N., 1995. Dynamic
Wrinkle Model in Facial Animation and Skin Aging.
In Visualization and Computer Animation, 6:165—
205, 1995.
Malik J. and Perona P., 1990. Pre-attentive discrimination
with early vision mechanism. In Journal of Optical
Society of America A, Vol.7, No.5, pp.923-932, 1990.
GRAPP 2006 - COMPUTER GRAPHICS THEORY AND APPLICATIONS
318
