Boosting face recognition speed with a novel divide-and-conquer approach

Abstract

Computational and storage space efficiencies of a novel approach based on appearance-based statistical methods for face recognition are studied. The new approach is a low-complexity divide-and-conquer method implemented as a multiple-classifier system. Appearance-based statistical algorithms are used for dimensionality reduction followed by distance-based classifiers. An appropriate classifier combination method is used to determine the resulting face recognized. FERET database and FERET Evaluation Methodology are used in all experimental evaluations. Time and space complexities of the proposed approach indicate that it, outperforms the holistic Principal Component Analysis, Linear Discriminant Analysis and Independent Component Analysis in computational and storage space efficiencies. The experimental results show that the proposed approach also provides better recognition performance on frontal images.