Discretized Gabor Statistical Models for Face Recognition
Jian Xin Zou, Chuancai Liu · International Journal of Digital Content Technology and its Applications · 2011
Aiming to the use of texture information of Gabor filtered face images, we present a new method matching multi-channel Gabor marginal statistical models for face recognition. Given partitions on channel Gabor magnitude spaces, the ensemble of magnitude sets of Gaborfaces is modeled as a probabilistic realization of a set of multinomial models. The histogram method is adopted to obtain corresponding empirical models for algorithmic implementation. With the Fisher geometry on multinomial family, the Fisher information distance is extended to the closure of each channel model space for quantifying information divergence between factorial histograms in a natural product framework. The root-mean-squared (RMS) extended Fisher information distance is used for product histogram match. The method is straightforward without learning step and limitation of all images in same size. Its effectiveness is demonstrated by the promising recognition results on ORL and Georgia Tech face databases.