IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, VOL. PAMI-3, NO. 2, MARCH 1981
Theoretical remarks and application," in Proc. 4th Conf. Inform.
Theory, Prague, 1965, pp. 635-660.
[201 Y. T. Chien and K. S. Fu, "Selection and ordering of feature
observations in a pattern recognition system," Inform. Contr.,
vol. 12, pp. 395-414, 1968.
[21] J. T. Tou and R. P. Heydorn, "Some approaches to optimum fea-
ture extraction," in Computer and Information Sciences II, J. T.
Tou, Ed. New York: Academic, 1967, pp. 41-122.
[22] T. Y. Young, "The reliability of linear feature extractors," IEEE
Trans. Comput., vol. C-20, pp. 967-971, 1971.
[23] T. M. Cover and P. E. Hart, "Nearest neighbor pattern classifica-
tion,"IEEE Trans. Inform. Theory, vol. IT-13, pp. 21-27, 1967.
[24] K. Kugunaga and D. L. Kessell, "Estimation of classification
error," IEEE Trans. Comput., vol. C-20, pp. 1521-15 27, 1971.
Tzay Y. Young (S'58-M'63-SM'80) received
the B.S. degree from the National Taiwan Uni-
versity, Taipei, Taiwan, China,
in
1955,
the
M.S.
degree from the University of Vermont, Bur-
lington, in 1959, and the Dr.Eng. degree from
the Johns
Hopkins University, Baltimore, MD,
in
1962, all in electrical
engineering.
From 1962 to 1963 he was a Research Asso-
i
ciate at
Carlyle Barton
Laboratory
of the Johns
Hopkins University, and from 1963 to 1964 he
was a member of the Technical staff of Bell
Telephone Laboratories, Murray Hill, NJ. He was on the faculty of
Carnegie-Mellon University from 1964 to 1974, and was on leave at
NASA Goddard Space Flight Center from 1972 to 1973. Since 1974
he has been a Professor of Electrical Engineering at the University of
Miami, Coral Gables, FL. He is coauthor (with T. W. Calvert) of
Classification, Estimation, and Pattern Recognition, published by
American-Elsevier. He was an Associate Editor of the IEEE TRANS-
ACTIONS ON COMPUTERS from 1974 to 1976, and is currently a member of
Editorial Committee of IEEE TRANSACTIONS PATTERN ANALYSIS AND
MACHINE INTELLIGENCE.
Philip S. Liu (S'70-M'75) was born in Wai
Chow, China, on November 19, 1945. He re-
ceived the B.S. degree in electrical engineering
from the University of Wisconsin, Madison, in
1970, and the M.S. and Ph.D. degrees in electri-
cal engineering from Purdue University, West
Lafayette, IN, in 1972 and 1975, respectively.
000--i--0X_ <--;
He
joined
the
faculty
of the
University
of
Miami, Coral Gables, FL, in 1975, and he is
currently Associate Professor of Electrical Engi-
neering. His current research interests include
database systems and computer architecture.
Dr. Liu is a member of the Association for Computing Machinery and
Eta Kappa Nu.
Romulo J. Rondon was born in Caripito, Vene-
zuela,
on November
24,
1950. He received the
degree from the University of Carabobo in
electrical engineering in 1974 and the M.S.
i%0S
+? t00i
ledegree from the University of Miami, Coral
Gables, FL, in 1978.
He held research
assistantships during his
\ j g 4- t i0-; -raduate work at the University of Miami where
> i m he was engaged in a research in data compres-
sion and pattern recognition. He is currently
with the Engineering Department of the Ford
Motor Company of Venezuela.
Mr. Rond6n is a member of Eta Kappa Nu.
An Approximate Solution to Normal Mixture
Identification with Application to Unsupervised
Pattern Classification
JACK-GERARD POSTAIRE, MEMBER, IEEE, AND CHRISTIAN P. A. VASSEUR
Abstract-In this paper, an approach to unsupervised pattern classifi-
cation is discussed. The classification scheme is based on an approxima-
fton of the probability densities of each class under the assumption that
the input patterns are of a normal mixture.
The proposed technique for identifying the mixture does not require
prior information. The description of the mixture in terms of convex-
ity allows to determine, from a totally unlabeled set of samples, the
number of components and, for each of them, approximate values of
the mean vector, the covariance matrix, and the a priori probability.
Manuscript received November 26, 1979; revised April 28, 1980.
J.-G. Postaire is with the Laboratoire d'Electronique et d'Etude des
Systemes Automatiques, Faculte des Sciences, Rabat, Morocco.
C. P. A. Vasseur is with the Centre d'Automatique, Universite de
Lille, Villeneuve d'Ascq, Cedex, France.
Discriminant functions can then be constructed. Computer simula-
tions show that the procedure yields decision rules whose performances
remain close to the optimum Bayes minimum error-rate, while involving
only a small amount of computation.
Index Terms-Convexity, minimum error-rate classifilcation, normal
mixture identification, unsupervised classification.
SUMMARY
IN THIS WORK, we explore a new approach to the Gaussian
multicategory classification problem, in an unsupervised
environment. All the statistics of each class, as well as the
number of classes, are unknown. An approximate solution to
0162-8828/81/0300-0163$00.75
© 1981 IEEE
163