Gabor Texture Information for Face Recognition Using the Generalized Gaussian Model

Abstract

To reduce the dimensionality of the Gabor feature, this paper explores texture information from Gabor coefficients and presents two kinds of new Gabor texture representations for face recognition: Gabor real part-based texture representation (GRTR) and Gabor imaginary part-based texture representation (GITR). Specifically, GRTR and GITR are obtained using the generalized Gaussian distribution (GGD) to model the real and imaginary parts of Gabor coefficients, respectively. The estimated model parameters serve as texture representation. Experiments performed on Yale and FERET databases show that the proposed texture representations GRTR and GITR significantly outperform the widely used Gabor magnitude in terms of recognition accuracy.