Coding 3D Gabor Features for Hyperspectral Palmprint Recognition

Compared to the fruitful research outputs in 2D palm print recognition, the research in hyper spectral palm print recognition is quite limited in literature. When 2D slices of hyper spectral data was processed separately and then fused at different levels for palm recognition, the information contained in the 3D data is not fully exploited. We proposed a 3D Gabor wavelet based approach in this paper to extract features in spatial and spectrum domain simultaneously. A set of 3D Gabor wavelets with different frequencies and orientations were designed and convolved with the cube to extract discriminative information in the joint spatial-spectral domain. For each location in the 3D cube, the wavelet who produces the maximum response is identified and the response is coded using a two-bits code according to the phase information. The similarity between two hyper spectal cubes are then calculated using hamming distance measurement. The HK-PolyU Hyper spectral Palm print Database captured from 380 palms were used for experiments. Results show that the fused feature substantially outperformed the accuracy of individual wavelet. As low as 4% EER was achieved.