A Semi-Supervised Sparsity Discriminant Analysis Algorithm for Feature Extraction
Hui Sun, Yugen Yi, Ying Hua Lv, Hui Tao Cai, Jianzhong Wang · Advanced materials research · 2012
Recently, l1-graph was proposed as a new graph construction procedure. Compared with the kNN-graph and ε-graph, l1-graph possesses three advantages: robustness to data noise, sparsity and datum-adaptive neighborhood selection. In this paper, we propose a novel semi-supervised feature extraction method based on l1-graph termed Semi-supervised Sparsity Discriminant Analysis (S3DA). The proposed S3DA maintains the advantages of l1-graph, and more importantly, it has better capacity of discrimination for classification. Experimental results on face and gene expression databases demonstrate our proposed approach outperform some other state of the art algorithms, and also show the feasibility and effectiveness of our proposed approach.