Robust face recognition using wavelet transform and autoassociative neural network

In this paper, an efficient face recognition system using wavelet transform (WT) and modular autoassociative neural network (AANN) is proposed. WT, which has superior feature representation capability in multiresolution space and also less sensitive to noise and variation to lighting condition, is used to extract the features. The AANN which perform identity mapping of input space is used to capture the distribution of the low resolution face data obtained from WT. To avoid over fitting, over training and small-sample effect problem, we construct separate AANN for each person. To evaluate the proposed scheme, experiments have been conducted using ORL database and Yale A database for three cases namely normal images, noisy images and occluded images. In all the three cases, the modular AANN scheme produces better recognition rate compared to PCA, LDA and kernel associative memory (KAM). In particular, the proposed method outperforms the other methods in the case of occluded images.