An Improved Fingerprint Recognition Algorithm Using EBFNN

Radial basis function neural network (RBFNN) is widely applied in pattern recognition. Comparing with the Gaussian function of RBFNN, the ellipsoidal basis function (EBF) can make the partition of input space more specific. Ellipsoidal basis function neural network (EBFNN) is a forward-feedback neural network of which hidden-layer function is EBF, which can make the partition of input space limitary and bounded. A novel fingerprint recognition algorithm using EBFNN has been proposed in this paper. Wavelet transform (WT) is a mathematical tool based on many-layer function decomposition. Firstly, fingerprint features are extracted by six-layer WT decomposition on binary images. Then, the extracted features are input into the designed EBFNN. Finally, EBFNN is trained and fingerprint recognition is accomplished by the trained EBFNN. The resulting algorithm is tested on three different fingerprint verification challenge datasets and shows the algorithm has higher recognition rate and speed than WT-RBF.