关键词: 降维, 连接性与权重, 稀疏表示, 边约束

Abstract: A Semi-Supervised Dimensionality Reduction method based on Sparsity Representation（SpSSDR） is proposed.Unlike other semi-supervised dimensionality reduction methods that construct graphs in steps，SpSSDR simultaneously defines the connectivity and the edges’ weights of a graph via sparsity reconstruction coefficients，and then exploits pairwise constraints for dimensionality reduction.Experiments on high dimensional facial data show that SpSSDR is not only robust to noise but also making use of pairwise constraints efficiently.

Key words: dimensionality reduction, connectivity and weights, sparsity representation, pairwise constraints