, Volume 23, Issue 4, pp 341–350 | Cite as

Modification of Karhunen-Loeve transform for pattern recognition



Application of the Karhunen-Loeve transform in pattern recognition problems necessitates knowledge of the mean vectors for the training sets of different classes for construction of the respective covariance matrices. This modality poses problems for recognizing an unknown signal in a multi-class environment and for its on-line implementation. This is becausea priori it is unknown which mean vector of the several classes is to be subtracted from an unknown input signal. To remove this difficulty a global mean approach has been proposed. The proposed method has been applied to synthetic and experimentally observed acoustic signals successfully and its performance with respect to pattern recognition and data compression has been compared with the conventional method.


Karhunen-Loeve transform global mean approach synthetic signal acoustic signal 


Unable to display preview. Download preview PDF.

Unable to display preview. Download preview PDF.


  1. Ahamed N, Rao K R 1975Orthogonal transforms for digital signal processing (Berlin: Springer-Verlag)Google Scholar
  2. Bhama S, Singh H, Phadte N D 1993 Parallelism for the faster implementation of the K-L transform for image compression.Pattern Recog. Lett. 14: 651–659CrossRefGoogle Scholar
  3. Chien Y T, Fu K S 1967 On the generalized Karhunen-Loeve expansion.IEEE Trans. Inf. Theory IT-13: 518–520CrossRefGoogle Scholar
  4. Chien Y T, Fu K S 1968 Selection and ordering of feature observations in a pattern recognition system.Inf. Control 12: 395–414CrossRefGoogle Scholar
  5. Fu K S 1968Sequential methods in pattern recognition and machine learning (New York: Academic Press)MATHGoogle Scholar
  6. Fukunaga K 1972Introduction to statistical pattern recognition (New York: Academic Press)Google Scholar
  7. Jain A K 1976 A fast Karhunen-Loeve transform for a class of random processes.IEEE Trans. Commun. COM-24: 1023–1029CrossRefGoogle Scholar
  8. Laaksonen J, Oja E 1996 Subspace dimension selection and averaged learning subspace method in handwritten digit classification.Proc. of the Int. Conf. on Artificial Neural Networks, Bochum, Germany, pp 227–232Google Scholar
  9. Laaksonen J 1997 Local subspace classifier and local subspace SOM.Proc. of the Workshop on Self-Organizing Maps, Otaniemi, Finland, pp 32–37Google Scholar
  10. Oja E 1983Subspace methods of pattern recognition (New York: Research Studies Press)Google Scholar
  11. Oja E 1989 Neural networks, principal components, and subspaces.Int. J. Neural Syst. 1: 61–68CrossRefMathSciNetGoogle Scholar
  12. Oja E 1992 Principal components, minor components, and linear neural networks.Neural Networks 5: 927–935CrossRefGoogle Scholar
  13. Oja E, Kärki T 1987 Properties of the ALSM subspace classifier for texture analysis.Proceedings of the 5th Scandinavian Conf. on Image Analysis, Stockholm, pp 419–426Google Scholar
  14. Oja E, Kohonen T 1988 The subspace learning algorithm as a formalism for pattern recognition and neural networks.IEEE Int. Conf. on Neural Networks, San Diego, California, pp I277–I284Google Scholar
  15. Oja E, Kuusela M 1983 The ALSM algorithm an improved subspace method of classification.Pattern Recogn. 16: 421–427CrossRefGoogle Scholar
  16. Rosenfeld A 1976Digital picture processing (New York: Academic Press)Google Scholar
  17. Tomita S, Noguchi S, Oizumi J 1970 On evaluation of hand-written characters by Karhunen-Loeve orthonormal system.Proc. 3rd. Hawaii Int. Conf. on Syst. Sci., Part I, pp 501–504Google Scholar
  18. Tou J T, Gonzalez R C 1974Pattern recognition principles (London: Addison-Wesley)MATHGoogle Scholar
  19. Watanabe S, Pakvasa N 1973 Subspace method of pattern recognition.Proceedings of the first Int. Conf. on Pattern Recognition, Washington, DC, pp 25–32Google Scholar

Copyright information

© Indian Academy of Science 1998

Authors and Affiliations

  1. 1.Variable Energy Cyclotron CentreCalcuttaIndia

Personalised recommendations