Block-wise two-directional 2DPCA with ensemble learning for face recognition

Two-Dimensional Principal Component Analysis (2DPCA) is a well-known feature extraction method for face recognition. One of the main drawbacks of this method, in comparison with the vector-based PCA, is that it needs many more coefficients to represent the feature matrix of an image. Two-Directional 2DPCA ((2D)^2PCA), proposed in the literature, attempts to alleviate this problem. However, it fails to improve the recognition accuracy of 2DPCA. In addition, (2D)^2PCA follows a global feature extraction approach that might fail to preserve some important local features. In this paper, we propose Block-Wise (2D)^2PCA to enhance the performance of (2D)^2PCA by preserving the local informative variations. On average, the feature matrices produced by the proposed method and those formed by (2D)^2PCA are about the same size. However, our experiments on four face recognition databases indicate that our method is superior to (2D)^2PCA in terms of the recognition accuracy.