We extensively investigate robust sparse two dimensional principal component analysis (RS2DPCA) that makes the best of semantic, structural information and suppresses outliers in this paper. The RS2DPCA combines the advantages of sparsity, 2D data format and L1-norm for data analysis. We also prove that RS2DPCA can offer a good solution of seeking spare 2D principal components. To verify the performance of RS2DPCA in object recognition, experiments are performed on three famous face databases, i.e. Yale, ORL, and FERET, and the experimental results show that the proposed RS2DPCA outperform the same class of algorithms for face recognition, such as robust sparse PCA, L1-norm-based 2DPCA.
Meng, Jicheng and Zheng, Xiaolong
"Robust Sparse 2D Principal Component Analysis for Object Recognition,"
Applied Mathematics & Information Sciences: Vol. 07:
6, Article 45.
Available at: https://digitalcommons.aaru.edu.jo/amis/vol07/iss6/45