be overcome by considering learning conditions such as 
size of the self-organizing maps, the learning coefficient, 
and kinds of images using learning and changing method 
of weights. These we will be undertake in future studies. 
6.3 Classifying Facial Expressions 
In this study, we analyzed changes in the victory neuron of 
the eyebrows and the eyes and the mouth in 3- 
dimensional space. Using this method, five kinds of facial 
expressions were classified out of the basic six types of 
facial expressions. Our analysis demonstrates that 
important elements of facial images are reflected in 3- 
dimensional information. Visualization was possible 
reducing the dimension of image information into 3- 
dimensional information. 
In the future, we will design other methods for analyzing 
the change of victory neuron. 
7. CONCLUTIONS 
In this study, we attempted to classify facial expressions by 
using self-organizing maps. By inputting various images 
of facial expressions into self-organizing maps and 
changing the interconnection weights, we were able to 
make self-organizing maps capable of classifying image 
features. This study demonstrates that by inputting 
images into self-organizing maps, it is possible to 
representing image features as a victory neuron number in 
self-organizing maps. Thus it is possible to consider 
changes in images as changes in the victory neuron 
number. 
Movements of facial segments have features peculiar to 
each facial expression. The victory neuron showed 
changes peculiar to each facial expression when image 
sequences of facial expressions were input into self- 
organizing maps. 
By analyzing the changes in the victory neuron, we were 
able to classify facial expressions, thus demonstrating the 
possibility of recognizing facial expressions by using this 
method. In the future, we will strive to develop an 
algorithm for recognizing facial expression and a method 
for automatically tracking such facial segments as the 
eyebrows and the eyes and the mouth. 
448 
References 
References from Journals: 
Hiroshi, K., Fumio, H., 1995. Monitoring of Facial 
Expressions. J.SICE, Vol.v34, No. 4, pp248-254. 
Hiroshi, Y., Psychological Model of Recognizing Facial 
Expressions. J.SICE, Vol.33, No. 12, pp-1063-1069. 
Kenji, M., 1991. Recognition of Facial Expression from 
Optical Flow. IEICE Transactions, Vol. E-74, No. 10, 
pp3474-3483. 
Noboru, S., 1994. Expectation by a Psychiatrist to 
Recognition Face Image. Medical Imaging Technology, Vol. 
12, No. 6, pp.700-709. 
Shigeo, M., 1994. Recognition of Facial Expression —From 
the View of Engineering-. Medical Imaging Technology, Vol. 
12, No. 6, pp688-693. 
References from Books: 
Paul, E., Wallace, V. F., Unmasking the Face. Printice-Hall. 
Teuvo, K., 1996. Self-organizing Maps. Springer-Verlag, 
Tokyo, pp102-171. 
References from Other Literature: 
Hitoshi, O., Changsuk, C., Haruyuki, M., 1994. Computer 
Recognition of Facial Expressions. Conference on Imaging 
Engineering, pp23-26. 
Hiroshi, K., Fumio, H., Susumu, Ll, 1993. Dynamic 
Recognition of 6 Basic Facial Expressions by Recurrent 
Neural Network. Technical Report of IEICE, HC92-59, 
pp11-16. 
Katsuhiro, M., Chil-Woo, L., Saburo, T., 1994. Lecture 
Notes in Computer Science, Vol.800, Computer Vision — 
ECCV’94, pp513-520. 
Tatsumi, S., 1995. Image Feature Extraction Using Wavelet 
Transformation and Its Application for Facial Expression 
Recognition. Technical Report of IEICE. IE94-147, pp15-22. 
  
FAC 
Abs 
We p 
can n 
is div 
mark 
motic 
direc 
repor 
Face and g 
because th 
is essentia 
machine. ] 
medical di 
lips, perio 
function. 
Most s! 
ers to sub 
surement 
tic points. 
be fixed t« 
These rest 
cial motio 
In this 
ture point 
In the rec 
points" ap 
marked in 
and used 
ages. 
We fol 
images of 
image seq 
sequence