Incremental Principal Component Analysis Based On Reduced Subspace Projection

Subspace projection (SP) is a kind of efficient subspace tracking algorithm, and it is an incremental principal component analysis algorithm too. In this paper the SP algorithm is first analyzed in detail; then, based on the eigenvector´s property the computation complexity of SP is reduced from O(N²(P+1)) to O(N²); finally, the covariance matrix is replaced with approximated covariance matrix which is composed of large eigenvalues and their corresponding eigenvectors, the computation complexity can be reduced to O(N(P+1)) further. Experiment results based on ORL face database demonstrate the efficiency of our proposed algorithm.