Abstract

When the statistics of pixels in an image can be described by the first-order Markov process, which implies decreasing correlation between pixels of increasing geometrical separation, the least-squares linear mapping technique (LSLMT) can be simplified for pattern recognition. The basis functions of LSLMT are shown to be a linear combination of the means of all training classes. Experimental results obtained by using the basis functions from the simplified method show reliable recognition.

© 1984 Optical Society of America

Optical implementation of the least-squares linear mapping technique for image classification

Zu-Han Gu, James R. Leger, and Sing H. Lee
J. Opt. Soc. Am. 72(6) 787-793 (1982)

Comparison of statistical pattern-recognition algorithms for hybrid processing. I. Linear-mapping algorithms

Q. Tian, Y. Fainman, Z. H. Gu, and Sing H. Lee
J. Opt. Soc. Am. A 5(10) 1655-1669 (1988)

Statistical recognition of color images

Zu-Han Gu, Sing H. Lee, and Y. Fainman
Appl. Opt. 26(15) 3145-3152 (1987)

Unified synthetic discriminant function computational formulation

David Casasent
Appl. Opt. 23(10) 1620-1627 (1984)

Hybrid optical–digital pattern recognition

Fritz Merkle and Thomas Lörch
Appl. Opt. 23(10) 1509-1516 (1984)