Palmprint Recognitionvia Bandlet, Ridgelet, Wavelet and Neural Network

Abstract

Palmprint recognition has emerged as a valid biometric based personal identification tool. Palmprints with high resolution features such minutia points, ridges and singular points or low resolution features such as wrinkles and principals determine their applications. In this paper a 700nm spectral band PolyU hyperspectral palmprint database is utilized and the multiscale band let image transform is utilized in features extraction; moreover, its results are compared with the ridgelet and 2D discrete wavelet results. The size of features is reduced using principle component analysis and linear discriminate analysis; in addition, a feed forward back-propagation neural network is used as a classifier. The results show that the recognition rate accuracy of the band let transform outperforms others.