|
Title:
|
SELECTION OF ORTHOGONAL FEATURES IN FISHER DISCRIMINANT ANALYSIS |
|
Author(s):
|
Zhaojia Sun , Miseon Choi , Cheong Hee Park , Young-kuk Kim |
|
ISBN:
|
978-972-8924-63-8 |
|
Editors:
|
Hans Weghorn and Ajith P. Abraham |
|
Year:
|
2008 |
|
Edition:
|
Single |
|
Keywords:
|
Feature Extraction; KFD; FLD; Quadratic Optimization Problem; Kernel |
|
Type:
|
Short Paper |
|
First Page:
|
102 |
|
Last Page:
|
106 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Conventional Fisher Linear Discriminant analysis (FLD) for pattern classification yields a single dominant feature which
comes from the unique eigenvector having a nonzero eigenvalue due to a rank-one matrix. Consequently T.Okada had
developed an optimal orthogonal system for FLD, and it had a good performance on applications. Recently S. Mika
extended FLD to nonlinear kernel space. The extension is called Kernel Fisher Discriminant (KFD) which is related to a
support vector machine. Despite of its better performance, it has the limitation which has only m-1 features on m-class
classification problem. We propose a novel method that applies on this orthogonal system and achieves better
performance by finding more orthogonal features in the nonlinear feature space. The simulation result shows that our
method is more effective in pattern recognition. |
|
|
|
|
|
|
