Fingerprint Recognition Using VNDLLE and GAEBFNN

摘要

Aiming at the nonlinear distortions in fingerprint images and the difficulties in parameters estimation of Ellipsoidal Basis Function Neural Network (EBFNN), a novel fingerprint recognition method is proposed by variable neighborhood discriminant locally linear embedding (VNDLLE) and genetic optimization EBFNN (GAEBFNN). Firstly, wavelet transform(WT) was applied on raw images. Then, the dimension of feature vector was reduced by VNDLLE. Finally, the parameters of EBFNN, such as weight, center and width, were estimated by GA and fingerprint recognition was realized by GAEBFNN. The proposed algorithm on NIST-4 has achieved the accuracy of 92.7% and average recognition time of 1.053 seconds.