Analysis of Subspace-based Face Recognition Techniques under Changes in Imaging Factors

The recognition accuracy of three face recognition algorithms based on subspace methods (i.e., the eigenfaces approach, the fisherfaces technique, and the probabilistic eigenfaces approach) is tested for different gallery sizes, image resolutions and number of bits per pixel. The same experiments are then repeated for various combinations of these techniques and the performance of these combinations with that of the individual methods is compared. The results reveal some interesting relations between recognition accuracy and image resolution and also indicate that the combinations of these techniques do yield some improvement in performance over the individual methods warranting that it would be worthwhile to further investigate the unification of these approaches