Fast and Accurate Face Recognition Using Support Vector Machines

The challenge of face recognition software is the rapid and accurate identification or classification of a query image, or set of query images, based on a set of known target images. Although Support Vector Machines (SVMs) are known to be accurate for the classification problem they are limited in this application by the time required for training which is dependent on the length of the feature vector. In this paper we present a novel method of feature reduction that greatly reduces computational time with minimal reductions in accuracy. It is shown that for Experiments 1, 2 and 4 of the the Face Recognition Grand Challenge Version 1, the feature reduction can make SVMs competitive with principal component analysis.