Hyperspectral Face Recognition with Spatial-Spectral Fusion Information and Gabor Feature

A hyperspectral face recognition was proposed using the HK-PolyU database in this paper. Twelve spectral bands were chosen from the hyperspectral face image data cubes according the spectral reflection property of skin, and then Gabor features were extracted for each selected spectral gray image respectively. Next, the feature fusion and decision fusion were studied. For the feature fusion, the fusion image was constructed by combining the twelve Gabor-feature vectors, and the weigh coefficients were decided by both the spectral reflection and the recognition accuracy. Maximum voting system was employed in the decision fusion. The validation experiment results show that, the recognition accuracy can reach to 96.5% when three image cubes are selected as training from each class. The recognition speed is more than 3 times of that without band selection. The extensive experiments show that the promising proposed approaches are superior both in the recognition accuracy and in recognition speed.