Abstract
This paper proposes a novel face recognition and representation approach using Gabor filters and Multi-band Gradient Component Pattern. The face image is first convolved with multi-scale, multi-orientation Gabor filters and Gabor volume is created using Gabor magnitudes and Gabor phases. Multi-band Gradient Component Pattern (MGCP) is used to extract geometry features in the orthogonal gradient space. Weighted Histogram Intersection is used in measuring dissimilarity between two face images. Conditional Mutual Information and Linear Discriminant Analysis are used for discriminant classification of feature space.