1. INTRODUCTION
Recognising faces affected by perturbations and changes in imaging conditions (illumination and/or blur) is very challenging because such effects not only degrade the quality of face, but also cause translation, rotation, occlusion and scale errors in the normalised face image. These challenges are manifested in large variability in facial appearance of the same person. The problem is aggravated by a high dimensionality of the face data and a small sample size.