Kernel Fisher discriminant analysis of Gabor features for online palmprint verification

We propose an online palmprint identification and verification algorithm with the use of kernel Fisher discriminant analysis (KFD) on the Gabor wavelet representation of palm images. Desirable palm features are derived by Gabor wavelets on the palm region. The KFD method is then employed to extract higher order relations among the Gabor-palm images for palmprint recognition. As a real-world application, the proposed algorithm was adapted into a novel online palmprint verification system that was employed in a student laboratory for 3 months. The feasibility of the Gabor-based KFD method was successfully tested on our proposed online palmprint system and on two data sets: KTU database, acquired in this real-life application, and the PolyU database. Comparing with existing PCA, KPCA, and Fisher discriminant analysis, the proposed method gives superior results on the KTU palmprint database. Furthermore, for palmprint recognition, our approach provides highly competitive performance (99.714% recognition rate and 0.078% equal error rate) with respect to the published palmprint recognition approaches tested with the same scenario on the public PolyU database.