back
United States Patent 5,774,576
United States Patent 5,774,576
Cox, et. al. Jun. 30, 1998
Pattern recognition by unsupervised metric learning
Inventors: Cox; Ingemar J. (Lawrenceville, NJ); Yianilos; Peter N. (Princeton, NJ).
Assignee: NEC Research Institute, Inc. (Princeton, NJ).
Appl. No.: 503,051
Filed: Jul. 17, 1995
Intl. Cl. : G06K 9/62
Current U.S. Cl.: 382/160; 382/118; 382/225; 382/228
Field of Search: 382/160, 159, 224, 228, 225, 118, 124, 155, 133, 165, 170, 187, 229; 395/2.49, 20, 21, 23; 364/275.9, 468.17; 358/467; 704/231, 232, 240, 243, 245, 251, 256
References Cited | [Referenced By]
U.S. Patent Documents
4,661,913Apr., 1987Wu et al. 364/500
5,075,896Dec., 1991Wilcox et al. 382/228
5,345,535Sept., 1994Doddington 395/2.45
5,461,699Oct., 1995Arbabi et al.
5,491,758Feb., 1996Bellegrada et al. 382/228

Other References

T. Kanade, "Picture Processing System by Computer Complex and Recognition of Human Faces," (Ph.D. thesis), Dept. Of Information Science, Kyoto University, Nov. 1973, pp. 75-91.

Satosi Watanabe, Frontiers of Pattern Recognition, Academic Press, 1972.

M. Bongard, Pattern Recognition, Spartan Books, 1970, ISBN: 0-87671-111-2.

Laveen N. Kanal, Pattern Recognition, Thompson Book Co., 1968.

Piper et al., Stein's Paradox and Improved Quadratic Descrimination of Real and Simulated Data by Covariance Weighting, IEEE Conference Article, 1994, pp. 529-532.

K. S. Fu, Digital Pattern Recognition, Springer-Verlag, 1980, ISBN: 3-540-10207-8; 0-387-10207-8.

K. S. Fu, Syntactic Pattern Reconigtion and Applications, Prentice-Hall Inc., 1982, ISBN: 0-13-880120-7.

George C. Cheng et al., Pictorial Pattern Recognition, Proceedings of Symposium on Automatic Photointerpretation, Thompson Book Co., 1968.

Tzay Y. Young, Handbook of Pattern Recognition and Image Processing, Academic Press, Inc., 1986, ISBN: 0-12-774560-2.


Primary Examiner: Boudreau; Leo
Assistant Examiner: Werner; Brian P.
Attorney, Agent or Firm: Feig; Philip J.
Abstract

A pattern recognition method uses unsupervised metric learning starting from a mixture of normal densities which explains well observed data. An improved decision rule is provided for selecting the reference database element most likely to correspond to a query.

8 Claims, 1 Drawing Figures

Links

back