Weighted matrix distance metric for face images classification

Abstract

This paper proposes a novel weighted distance metric based on 2D matrices rather than 1D vectors and the eigenvalues for face images classification and recognition. This distance is measured between two feature matrices obtained by two-dimensional principal component analysis (2DPCA) and two-dimensional linear discriminant analysis (2DLDA). The weights are the inverse of the eigenvalues of the total scatter matrix of face matrices sorted in decreasing order and the classification strategy adopted is the nearest neighbour algorithm. To test and evaluate the efficiency of the proposed distance metric, experiments were carried out using the international ORL face database. The experimental results show the high performance of the weighted matrix distance metric over the Yang and the Frobenius distances.