Application of the Karhunen-Loève Expansion to Feature Selection and Ordering
IEEE Transactions on Computers
| Hi-index | 14.99 |
The subspace method proposed by Watanabe et al. is an important method of feature extraction in pattern recognition. Some new results described here provide a formulation of the subspace method that is particularly easy to implement and overcomes the various numerical difficulties that were inherent in earlier implementations of the method. The present formulation is also important in that it provides the key to the relation of this method to other methods of feature extraction.