Efficient face recognition based on weighted matrix distance metrics and 2DPCA algorithm

In this paper, a new similarity measure is developed for human face recognition, namely, weighted matrix distance. The key difference between this metric and the standard distances is the use of matrices and weights rather than the vectors only. The two feature matrices are obtained by two-dimensional principal component analysis (2DPCA). The weights are the inverse of the eigenvalues sorted in decreasing order of the covariance matrix of all training face matrices. Experiments are performed under illumination and facial expression variations using four face image databases: ORL, Yale, PF01 and a subset of FERET. The results demonstrate the effectiveness of the proposed weighted matrix distances in 2DPCA face recognition over the standard matrix distance metrics: Yang, Frobenius and assembled matrix distance (AMD).