Article ID | Journal | Published Year | Pages | File Type |
---|---|---|---|---|
410821 | Neurocomputing | 2007 | 6 Pages |
In this paper, we propose a new feature extraction method—parameterized direct linear discriminant analysis (PD-LDA) for small sample size problems. Similar to direct LDA (D-LDA), PD-LDA is a modification of KLB (the Karhunen–Loève expansion based on the between-class scatter matrix). As an improvement of D-LDA and KLB, PD-LDA inherits two important advantages of them. That is, it can be directly applied to high-dimensional input spaces and implemented with great efficiency. Meanwhile, experimental results conducted on two benchmark face image databases, i.e., AR and FERET, demonstrate that PD-LDA is much more effective and robust than D-LDA. In addition, it outperforms state-of-the-art facial feature extraction methods such as KLB, eigenfaces, and Fisherfaces.