Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

Fractals, Vol. 10, No.

1 (2002) 47–52
c World Scientific Publishing Company

FRACTALS IN EMOTIONAL FACIAL


EXPRESSION RECOGNITION

TAKUMA TAKEHARA
Department of Psychology, Doshisha University
Kyoto 602-8580, Japan
takehara@psychology.doshisha.ac.jp
FUMIO OCHIAI
Department of Human and Cultural Studies
Tezukayama University, Nara 631-8501, Japan
NAOTO SUZUKI
Department of Psychology, Doshisha University
Kyoto 602-8580, Japan

Received June 5, 2001; Accepted September 7, 2001

Abstract
Following the Mandelbrot’s theory of fractals, many shapes and phenomena in nature have been
suggested to be fractal. Even animal behavior and human physiological responses can also be
represented as fractal. Here, we show the evidence that it is possible to apply the concept of
fractals even to the facial expression recognition, which is one of the most important parts of
human recognition. Rating data derived from judging morphed facial images were represented
in the two-dimensional psychological space by multidimensional scaling of four different scales.
The resultant perimeter of the structure of the emotion circumplex was fluctuated and was
judged to have a fractal dimension of 1.18. The smaller the unit of measurement, the longer
the length of the perimeter of the circumplex. In this study, we provide interdisciplinarily
important evidence of fractality through its application to facial expression recognition.

Keywords: Facial Expression Recognition; Morphing; Multidimensional Scaling.

47
48 T. Takehara et al.
Prototype Prototype
1. INTRODUCTION face face

Many studies have demonstrated that the structure


of emotional facial expression recognition can be
represented as a circumplex in the two-dimensional
psychological space, consisting of valence (pleasure- Morphed
displeasure) as one dimension and the level of face
arousal as the other.1 – 7 Further experiments on the Fig. 1 Schematic representation of a trajectory in a con-
circumplex structure of facial expression recogni- tinuum running from one prototype to another depicted in
tion showed a more detailed structure than before.6 the previous studies. The change is fluctuating rather than
According to the experiments, morphed facial im- continuous.
ages, lying between two prototype facial expressions
(e.g. Happiness and Surprise) and prototype faces melanogaster13 and the diving duration of whales14
were represented in the two-dimensional psycholog- both showed a temporal fractal. In human phys-
ical space in a roughly circular order. Interestingly, iology, the brain wave is proved to have a fractal
complex morphed facial images with more than one dimension,15 and moreover, the heart rate spectrum
emotional component showed a rather complex dis- also shows a fractal.16 However, to our best knowl-
tribution rather than simply between their proto- edge, we have not seen any studies that discussed
types corresponding to their morphing proportions. fractals in the complex world of human recognition.
Although the structure of emotional facial It is natural to assume that the human recognition
expression recognition has been represented in the system has a fractal structure, because behavior or
two-dimensional space, it was interesting to note a physiological response, which we can explain as
that facial images formed a roughly circular struc- fractal, can be modified by the recognition system.
ture. In other words, these experiments showed a The complex shape in the representation of facial
somewhat different distribution of facial images. In images in the two-dimensional space might be in-
addition, results from Takehara and Suzuki (1997)6 fluenced by fractals. The purpose of this study,
showed complex and wiggled arrangements between therefore, is to examine the possibility of a fractal
two prototype faces on a circumplex rather than structure underlying the facial expression recogni-
smooth and continuous distribution. Their results tion which is one of the most important parts of the
also showed that prototypes and morphed facial human recognition system. If the representation of
images were located in somewhat different loca- the perimeter of an emotion circumplex is fluctu-
tions. The fearful face, for example, was located ated by a fractal, it would have no characteristic
near the surprised face in one experiment and was length scale and it must, therefore, have a fractal
not in the other experiment even though the two dimension of D.
experimental conditions were the same. If the two-
dimensional psychological space is continuous and
2. METHOD
homogeneous, one would expect that the changes in
the representative trajectory running from one pro- 2.1 Experimental Conditions
totype to another around a circumplex would be
smooth, and geometric relationships between the Considering that eight facial expressions of Hap-
prototype faces would be mapped in the continuous piness, Calm, Sleepiness, Sadness, Anger, Fear,
space (see Fig. 1). But, why did we get such com- Surprise and Excitement were typical prototypes,
plex and fluctuated representations? This kind of we designed the perimeter of a circumplex varying
fluctuation may mask the possible fractality of the clockwise as shown in Fig. 2. To prepare different
shape. scales, the following four experimental conditions
Fractals are a concern of new geometry, whose were set: (i) 32 divided, (ii) 40 divided, (iii) 48
object is to describe a great variety of natural divided, and (iv) 64 divided conditions.
structures that have irregularities of various sizes.
Until recently, various shapes or surfaces in nature 2.2 Participants
have been explained as fractal in terms of their self-
similarity.8 – 12 Furthermore, it has been reported The participants for each of the four experimental
that the feeding behavioral pattern of Drosophila conditions were all undergraduate students: 23 for
Facial Expression Recognition 49

Surprise

Fear Excitement

Anger Happiness

Sadness Calm

Sleepiness
(a)

Happiness Calm

(b)

Fig. 2 (a) Schematic representation of eight prototype facial expressions on the perimeter of a circumplex. Note that the
distances between prototypes are not exactly the same. (b) An example of Happiness-Calm pair of morphing process and its
morphed images in condition (i). Physical increment is kept constant. The other pairs were carried out in the same way.

condition (i), 19 for (ii), 31 for (iii), and 28 for (iv). adjacent prototype face pairs positioned on the
They volunteered to be research participants to get perimeter in each of the four different scales using a
a credit for a psychology course. All had normal or morphing software (Morpher for WindowsTM ). In
corrected-to-normal vision. condition (i), each interval between adjacent pro-
totypes was divided into four segments and three
morphed facial images were generated. In total,
2.3 Stimuli
32 facial images (eight prototypes and 24 morphs)
Based on the arrangement illustrated in Fig. 2, were generated around a circumplex for condi-
morphed facial images were generated between all tion (i). Similar to this condition, morphed facial
50 T. Takehara et al.
1.5
images were generated for the other three con- Excitement
Surprise
ditions: 40 facial images (eight prototypes and
32 morphs) for condition (ii), 48 facial images 1.0
(eight prototypes and 40 morphs) for (iii), and 64 Fear
facial images (eight prototypes and 56 morphs) for 0.5
(iv). The particular advantage of the morphing is
that it is possible to create highly objective facial Happiness
images and easy to control the amount of physi- 0.0

cal changes. Since the morphing has been used in Anger


many studies,17 – 20 its procedure would provide an -0.5
extremely useful tool to investigate the underlying Calm
structure of facial expression recognition. Sadness
-1.0

2.4 Procedure
-1.5
For each of the four experimental conditions, Sleepiness
participants were asked to rate all facial images on -2.0
a six-point Likert-type scale (from “not at all” to -2.0 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5
“very so much”) about each item of the eight pro-
totype emotions of Happiness, Calm, Sleepiness, Fig. 3 Configuration of prototype and morphed facial
images in the two-dimensional psychological space. This ex-
Sadness, Anger, Fear, Surprise and Excitement.
ample illustrates a resultant configuration of condition (ii).
In rating, facial stimuli were presented one at a Black rectangles indicate the prototypes and white rectan-
time in a random order on a 14.1-inch LCD with gles morphs. Adjacent faces are tied with a straight line.
240×320 pixels. The distance between a participant All morphs are arranged between their prototypes, but the
and a display was about 50 cm. Data were collected perimeter of a circumplex seems to be complex and wiggled
rather than smooth. The other conditions showed the same
by a Compaq personal computer (DESKPRO;
tendency.
Pentium II Processor 350 MHz) with a two-button
mouse and by a rating software made by an Inprise
C++ compiler (Borland C++ Builder Version 4).
No feedback was given. prototypes fell in a roughly circular order and were
rated in the same order under all conditions,3,4,6,7,22
and all morphs were plotted between their parent
3. RESULTS prototypes corresponding to their morphing pro-
portions. The resultant configuration in condition
In order to visualize the assumed two-dimensional (ii) is shown in Fig. 3. Roughly speaking, each
psychological space framework, we need to spec- configuration may appear to form a circle, but no-
ify the relative positions of faces within the psy- tice that the perimeter is complex and somewhat
chological space. This can be accomplished with wiggled rather than smooth. The distance of the
multidimensional scaling (MDS), which derives a perimeter of a circumplex is defined here as the to-
spatial representation of the stimuli from psycholog- tal distance, which is the sum of distances between
ical data, such as ratings of perceived similarity.21 adjacent facial images.
For each of the four experimental conditions, rating If the distance of the perimeter were exactly the
data were used to calculate a matrix of dissimilari- same for the four conditions despite the different
ties among eight facial expression categories, which scales, it would not have fractal properties, suggest-
was then subjected to MDS to find coordinates of ing that the perimeter has a fixed and characteristic
all facial images in two-dimensional space. For the length scale like a complete circle. On the contrary,
two-dimensional configuration, all RSQ values were if it were revealed that as the scale was made smaller
sufficiently high: 0.978 for condition (i), 0.971 for and smaller, the distance of the perimeter tended to
(ii), 0.976 for (iii), and 0.977 for (iv). RSQ is a mea- increase steadily without bounds, we would obtain
sure of the proportion of the variance in the data a fractal dimension, indicating that the perimeter
accounted for by the MDS solution. Similar to the had no characteristic length scale, and was fluc-
results of many previous studies, the positions of the tuated by a fractal dimension. To test this, we
Facial Expression Recognition 51

1.9 recognition was a circumplex,1 – 7,22 but no study has


stated that the perimeter was complex and wiggled
rather than smooth. Focusing on the subtle fluc-
1.8 tuation around the perimeter of a circumplex, we
examined whether the emotion circumplex had a
Log number of steps

fractal structure or not in four different scales. The


1.7 perimeter of a circumplex represented in the two-
dimensional psychological space showed a fractal
structure, indicating that human facial expression
1.6 recognition, in some sense, had fractal properties
and that the two-dimensional space itself was not
homogeneous everywhere. Why is the perimeter of
1.5 D = 1.18 a circumplex fluctuated? Let us discuss possible
reasons and the future application.
In relation to the complex fluctuation of the
-0.8 -0.7 -0.6 -0.5 -0.4 perimeter around a circumplex, can we say that the
Log step-length fluctuation occurs in a random fashion? The answer
Fig. 4 Linear regression of log distance of the perimeter is no, because if the wiggled perimeter were due to
on log step-length. The slope of the straight line gives −D, the randomness, the log/log plot in Fig. 4 could not
from which we find a fractal dimension D = 1.18. R2 value be linear. The log/log plot in this study showed a
is 0.998. linear relation. It was, therefore, suggested that the
fluctuated perimeter was not due to the randomness
but to a fractal. The perimeter, however, seemed
calculated the length of the perimeter for each of the
to be wiggled irregularly, the fractal regularity was
four experimental conditions using the coordinates
hidden in the structure.
in two dimensions.
One reasonable interpretation for the fluctuated
The relationship between fractal dimension D,
perimeter is that it may enhance the recognition
step-length r and the perceived length of the line
efficiency for complex facial expressions which have
l(r) measured with step r is given by the following:
various emotional components. For example, con-
l(r) ∝ r (1−D) . (1) sider one continuum around the perimeter run-
ning from Happiness to Excitement. These two
According to the definition by Mandelbrot (1982),10 emotion categories occupy the similar region in the
fractal dimension D must lie in the range 1 ≤ D ≤ 2
two-dimensional space. In addition, the morphs
for a line. The slope of the resulting line on the
generated by interpolation of these two prototypes
log/log plot shown in Fig. 4 was estimated by re-
are complex facial expressions which have at least
gression analysis, then a fractal dimension of 1.18
more than two emotional components — Happiness
was derived with an R2 value of 0.998, indicat-
and Excitement. The fact that the trajectory of
ing that a circumplex did not have a characteris-
the continuum was not linear but fractal might in-
tic length scale. The fact that the estimation of
dicate the enhancement of the recognition efficiency
D is a fraction, 1.18, is highly impressive because
for complex and subtle facial expressions through
the perimeter of the emotion circumplex in the psy-
some fluctuations.
chological space has not been considered to have a
The fact that the recognition system has
fractional dimension, or an integer dimension (two-
fractal properties can perhaps be applied not only
or three-dimension) has been major in psychology.
to psychology but also to other various domains
Furthermore, a fractional dimension itself never has
of research dealing with recognition. To name a
been conceptualized in the domain of the human
few: clinical psychology, developmental psychology,
recognition, either.
artificial intelligence, human communication, brain
research and recognition engineering. In the future,
it is hoped that theoretical and empirical verifica-
4. DISCUSSION
tions will be carried out in an interdisciplinary man-
Many studies have suggested that the representa- ner since the topic has important implications to
tion of the structure of emotional facial expression many disciplines.
52 T. Takehara et al.

ACKNOWLEDGMENTS 11. B. B. Mandelbrot, D. E. Passoja and A. J. Paullay,


Nature 308, 721 (1984).
We thank Professor James A. Russell for provid- 12. D. R. Morse, J. H. Lawton, M. M. Dodson and M.
ing us the facial stimuli and Shizuhiko Nishisato, H. Williamson, Nature 314, 731 (1985).
Kazuyasu Shigemoto and Hiroshi Watanabe for 13. I. Shimada, Y. Minesaki and H. Hara, J. Ethol. 13,
useful commentary. 153 (1995).
14. B. J. West, Fractal Physiology and Chaos in
Medicine (World Scientific, Singapore, 1990).
REFERENCES 15. Y. Shinagawa, K. Kawano, H. Matsuda, H. Seno and
H. Koito, Forma. 6, 205 (1991).
1. R. Adolphs, D. Tranel, H. Damasio and A. Damasio,
16. Y. Yamamoto and R. L. Hughson, Physica D68, 250
Nature 372, 669 (1994).
(1993).
2. R. Adolphs, D. Tranel, H. Damasio and A. Damasio,
17. A. J. Calder, A. W. Young, D. I. Perrett, N. L. Etcoff
J. Neurosci. 15, 5879 (1995).
and D. Rowland, Vistual Cog. 3, 81 (1996).
3. J. A. Russell and M. Bullock, J. Pers. Soc. Psychol.
18. A. J. Calder, D. Rowland, A. W. Young, I. Nimmo-
48, 1290 (1985).
Smith, J. Keane and D. I. Perrett, Cognition 76, 105
4. J. A. Russell and M. Bullock, Dev. Psychol. 22, 97
(2000).
(1986).
19. N. L. Etcoff and J. J. Magee, Cognition 44, 227
5. H. Schlosberg, J. Exp. Psychol. 44, 229 (1952).
(1992).
6. T. Takehara and N. Suzuki, Percept. Mot. Skills 85,
20. A. W. Young, D. Rowland, A. J. Calder, N. L.
1003 (1997).
Etcoff, A. Seth and D. I. Perrett, Cognition 63, 271
7. T. Takehara and N. Suzuki, North Am. J. Psychol.
(1997).
3, 217 (2001).
21. R. N. Shepard, Science 210, 390 (1980).
8. D. Avnir and D. Farin, Nature 308, 261 (1984).
22. M. Katsikitis, Perception 26, 613 (1997).
9. B. B. Mandelbrot, Science 156, 636 (1967).
10. B. B. Mandelbrot, The Fractal Geometry of Nature
(Freeman, New York, 1982).

You might also like