Extended Discriminant Nearest Feature Line Analysis for Feature Extraction

In this paper, a novel feature extraction algo-rithm, entitled Extended Discriminant Feature Line Analysis (EDFLA), is proposed. EDFLA is a Nearest Feature Line (NFL) metric based dimensionality reduction method. For small size sample problem, the existing prototype samples usually are not enough to describe the corresponding class. To extend the representation ability of the prototype sample set, a novel prototype sample set will be generated in EDFLA using the original prototype samples and NFL. EDFLA aims at minimizing the within-class scatter and maximizing the between class scatter of the novel generated prototype sample set. The experimental results on OIL20 image database and AR face database confirm the effectiveness of the proposed algorithm.