Neighbor graph construction based on sparse representation for neighborhood preserving embedding

作者：Li Bin; Du Anan; Zhang Yecheng; Yu Xiangchun; Yu Zhezhou^*

来源：Journal of Information and Computational Science, 2014, 11(14): 5019-5026.

DOI：10.12733/jics20104578

• Dimensionality reduction
• Dimensionality reduction algorithms
• Graph construction
• Neighbor graph
• Neighborhood preserving embedding
• Non-uniform illumination
• Sparse representation
• Yale face database

Neighborhood Preserving Embedding (NPE) which has been successfully applied in face recognition is a graph based dimensionality reduction algorithm. Neighbor graph plays a significant role on the performance of NPE. Face recognition which is a challenging task since it may encounter occlusion and non-uniform illumination effects. In this paper, a new neighbor graph construction algorithm for NPE is proposed. The new neighbor graph construction algorithm is based on sparse representation which is robust to occlusion and illumination. So, we name our new neighbor graph construction algorithm as Sparse Representation (SR) based neighbor graph construction algorithm, and name the neighbor graph constructed by SR as Sparse Representation based neighbor Graph (SRG). Then, we incorporate SRG into NPE, and we call this improved NPE as SRG-NPE. Finally, several experiments were conducted on three well-known face databases: AR, ORL, and YALE face database, and achieved satisfactory experimental results, especially in the case of occlusion.

• 出版日期2014