Classification Of Human Faces Using Karhunen-LoÉve Decomposition And Radial Basis Function Neural Networks
Nejib Smaoui, Matar Ibrahim Matar · International Journal of Computer Mathematics · 2003
A novel approach to categorize and identify human faces is presented. The approach is based on two techniques, namely Karhunen-Loéve decomposition (K-L) and radial basis function networks (RBF). K-L decomposition, known for its wide applications in scientific problems for data compression and feature identification, is used to extract coherent structures or eigenfunctions from two male and female subsets consisting of 63 faces each. To each eigenfunction out of the 52 eigenfunctions extracted from each subset, an energy percentage is assigned based on the eigenfunctions associated eigenvalues. By sorting the eigenvalues from largest to smallest an ordering of the eigenfunctions from most to least energetic is accomplished. We show that the first most energetic eigenfunction is enough to categorize a face. To identify a face, we project the set of faces onto the eigenfunctions and obtain a set of data coefficients. Then, we use the data coefficients scaled with their associated energies to show that by incorporating the energies better identification results are obtained. Furthermore, we use the data coefficients as input to a RBF neural network and show that the results are comparable to previously obtained ones.