Design and Implementation of a Face Recognition System Using Fast PCA

High speed security and defense applications demand a quick decision for face recognition which requires a computationally time-efficient algorithm. These algorithms are primarily used to generate egien values. The generation of eigen values by employing decomposition method normally provides solution in O(n3) time whereas an orthogonalizational process, called fast principal component analysis (PCA) provides the same in O(n²) time. However, because of an orthonormalization convergence condition of Grams-Schmidt (GS) iterative process, fast PCA could result in non-deterministic state, especially for high resolution images. This could be associated with orthogonal vector space in GS, which causes nonconvergence of eigen solution under limited iteration. A modification has been proposed in fast PCA to generate eigen values for images including those at high resolution. By using these generated eigen values, an algorithm has been developed to optimize the error rate in face recognition systems under varying dimensionalities. The developed technique which provides deterministic, time efficient and low error rate solution could be a useful tool for high speed image recognition systems.