Weighted Kernel Locally Linear Embedding for Dimensionality Reduction

摘要

A novel dimensionality reduction method following weighted kernel locally linear embedding was proposed. Regarding manifold learning as the theory foundation, the traditional locally linear embedding was extended to kernel space, thus the nonlinear data can be processed. By introducing the importance score, the proposed approach reduced the undesirable effect resulted by the noisy points and improved the classification performance. The COIL-20 (Columbia Object Image Library) database and ORL (Olivetti Research Laboratory) face database were used to test the novel algorithm, experimental results showed the proposed method can give better results in manifold learning and dimensionality reduction, not only improved the classification accuracy, but also achieved high stability because of its non-sensitive to noise.