Face recognition based on the uncorrelated discriminant transformation

The extraction of discriminant features is the most fundamental and important problem in face recognition. This paper presents a method to extract optimal discriminant features for face images by using the uncorrelated discriminant transformation and It L expansion. Experiments on the ORL database and the NUST603 database have been performed. Experimental results show that the uncorrelated discriminant transformation is superior to the Foley-Sammon discriminant transformation and the new method to extract uncorrelated discriminant features for face images is very effective. An error late of 2.5% ig obtained with the experiments on the ORL database. An average error rate of 1.2% is obtained with the experiments on the NUST603 database. Experiments show that by extracting uncorrelated discriminant features, face recognition could be performed with higher accuracy on lower than 16 x 16 resolution mosaic images. It is suggested that for the uncorrelated discriminant transformation, the optimal face image resolution can be regarded as the resolution m x n which makes the dimensionality N = mn of the original image vector space be larger and closer to the number of known-face classes. (C) 2001 pattern Recognition Society. Published by Elsevier Science Ltd. All rights reserved.