Sparse representation-based neighbor graph construction method for locality preserving projection

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

来源：Journal of Computational Information Systems, 2014, 10(12): 5219-5226.

DOI：10.12733/jcis10666

• Dimensionality reduction
• Dimensionality reduction method
• Graph construction
• High dimensional data
• Locality preserving projections
• Non-uniform illumination
• Sparse representation
• Strong robustness

Graph-based dimensionality reduction method is the popular approach for understanding high-dimensional data, such as the face image. However, the recognition rate will be severely affected due to occlusion and non-uniform illumination effects. Neighbor graph construction method plays a crucial role in the graph-based dimensionality reduction method. Therefore, building a good neighbor graph is very important. Sparse representation is applied to face recognition because of its strong robustness to occlusion and illumination. In this paper, we propose a new neighbor graph construction method based on sparse representation, and combine this method with locality preserving projection (LPP) algorithm. We call this improved algorithm as sparse representation-based neighbor graph construction method for LPP (SRB-LPP). Then, we do several experiments on three well-known face databases. The experiments show that our method achieved better recognition rates than the traditional neighbor graph construction method, especially in the case of occlusion.

• 出版日期2014