Robust two-dimensional principle component analysis

Two-dimensional principle component analysis (2DPCA) is a popular and fast feature extraction method. However, it doesn’t consider the outliers of the training samples. To address this problem, we present robust two-dimensional principle component analysis algorithm (R2DPCA), which gives a new weighted method for the evaluation of the total squared error. The solution for R2DPCA is also given. The proposed method is tested on AR face database and ORL face database, and the experimental results indicate that it is more effective than two-dimensional principle component analysis (2DPCA).