The Phase-based Gabor Fisher Classifier and its application to face recognition under varying illumination conditions

Abstract

The paper introduces a feature extraction technique for face recognition called the Phase-based Gabor Fisher Classifier (PBGFC). The PBGFC method constructs an augmented feature vector which encompasses Gabor-phase information derived from a novel representation of face images - the oriented Gabor phase congruency image (OGPCI) - and then applies linear discriminant analysis to the augmented feature vector to reduce its dimensionality. The feasibility of the proposed method was assessed in a series of face verification experiments performed on the XM2VTS database. The experimental results show that the PBGFC method performs better than other popular feature extraction techniques such as principal component analysis (PCA), the Fisherface method or the DCT-mod2 approach, while it ensures similar verification performance as the established Gabor Fisher Classifier (GFC). The results also show that the proposed Phase-based Gabor Fisher Classifier performs the best among the tested methods when severe illumination changes are introduced to the face images.