AN OPTIMIZED FACE RECOGNITION SYSTEM USING STABLE ORTHOGONALIZATION

This article presents a new architecture to implement an optimized face recognition system based on reduced dimensionality of covariance matrix. The leading components of reduced matrix are computed by using modified Gram-Schmidt orthogonalization (MGSO). Use of MGSO in fast principle component analysis has improved the convergence and accuracy of a face recognition system especially for high dimensional images. To optimize the performance of a face recognition system, receiver operating characteristics are simulated with least square fitting. Convex optimization is applied for optimum selection of system variables. It is demonstrated that the proposed technique, compared with decomposition, provides better discriminating power in Eigen space.