Dimension reduction is an important data preparation step for face recognition. A new nonlinear dimensionality reduction method called kernel neighborhood preserving embedding (KNPE) is proposed in this paper. This new method extends the well-known neighborhood preserving embedding (NPE) from linear domain to a nonlinear domain with the kernel trick that has been used kernel-based learning algorithms. Extensive experiments have been conducted on the three well-known face databases. The experimental results show that our proposed KNPE algorithm yields much better performance than the other related algorithms.
展开▼
