A fast and robust Gabor feature based method for face recognition

This paper describes a fast and robust Gabor feature based approach for face recognition. The most discriminative Gabor features are selected by the AdaBoost procedure, which are then subjected to the Generalized Discriminant Analysis (GDA) process for further class separability enhancement. Compared with the huge number of features used by typical classification algorithms using Gabor filters, our method needs only two hundred Gabor features. Whilst significant memory and computation cost has been reduced, our method still achieves very high recognition accuracy. 600 frontal facial images from the FERET database, with expression and illumination variations, are used to test the system. Only 4 seconds are required to recognize 200 face images, 97% accuracy has been achieved.