A Weighted Semi-Supervised Dimensionality Reduction Method Based Locally Linear Graph

摘要

In recent years, dimensionality reduction has been widely used in image processing, pattern recognition, and other related fields, and achieved a good performance. Especially supervised dimensionality reduction algorithms have been successfully applied to preprocess of data in classification. However, in many practical applications, the size of labeled sample is generally too small for traditional supervised dimensionality reduction algorithms (such as Linear Discriminant Analysis, LDA) to overcome the Small Sample Size (SSS) problem; moreover, a small amount of labeled samples usually lead the projected data into an inappropriate low-dimensional subspace. In order to solve the two problems above, this paper proposes a Weighted Semi-supervised Maximum Margin Criterion (WSMMC) algorithm. Our method utilizes the Locally Linear Embedding Graph to characterize the spatial distribution of non-labeled samples and avoid the SSS problem by taking advantage of MMC. Experimental results show that the algorithm is effective.