Part of Advances in Neural Information Processing Systems 16 (NIPS 2003)
Haifeng Li, Tao Jiang, Keshu Zhang
A new feature extraction criterion, maximum margin criterion (MMC), is proposed in this paper. This new criterion is general in the sense that, when combined with a suitable constraint, it can actually give rise to the most popular feature extractor in the literature, linear discriminate analysis (LDA). We derive a new feature extractor based on MMC using a different constraint that does not depend on the nonsingularity of the within-class scatter matrix Sw. Such a dependence is a major drawback of LDA especially when the sample size is small. The kernelized (nonlin- ear) counterpart of this linear feature extractor is also established in this paper. Our preliminary experimental results on face images demonstrate that the new feature extractors are efficient and stable.