Class-Specific Subspace-Based Two-Dimensional Principal Component Analysis for Face Recognition

In this paper, we proposed a class-specific subspace-based two-dimensional principal component analysis (2DPCA) for face recognition. In 2DPCA, 2D face image matrices do not need to be previously transformed into a vector. In this way, the spatial information can be preserved. Moreover, 2DPCA can achieve higher performance than PCA both in face recognition and face representation task. However, both PCA and 2DPCA are unsupervised techniques, no information of class labels are considered. Therefore, the directions that maximize the scatter of the data might not be as adequate to discriminate between classes. In recognition task, a projection is required to emphasize the discrimination between classes. The face-specific subspace (FSS) was proposed in concept of class-specific subspace. Each subspaces learned from the training images which correspond to only one class, thus the number of these subspaces is equal to the number of classes. Since the information of class labels are considered in FSS, so the discriminant power can be improved. We apply 2DPCA to class-specific concept in our framework which consists of two methods: the first one, we apply FSS to 2DPCA method and the second one, we use the bilateral-projection-based 2DPCA (B2DPCA) instead of 2DPCA. The B2DPCA does not only allow further reducing of the dimension of feature matrix of 2DPCA-based but also improving the classification accuracy. Experimental results on Yale face database showed an improvement of our proposed techniques over the conventional 2DPCA