Investigation of Dimensionality Reduction on Numerical Attribute Features in a Finger Vein Identification System

With the large number of people travelling internationally, there is an increasing demand to be able to deal with security clearance rapidly and with a minimum of inconvenience. Using finger vein biometric traits fulfils these requirements. In previously-reported work, the data obtained from finger veins underwent dimensionality reduction using principal components analysis (PCA) followed by linear discriminant analysis (LDA) and this was shown to improve the identification rate compared to the more commonly applied Discrete Wavelet Transform (DWT). Although PCA was found to be effective at reducing the noise residing in the discarded dimension, this work demonstrates that the corresponding eigenvalue may in fact also contain useful local information that is important in identification and so should be retained. To overcome this problem, this paper proposes the use of feature extraction using DWT and local binary patterns (LBPs) to generate the feature vectors, before they undergo dimensionality reduction using PCA. Support Vector Machines (SVMs) are used for classification. The performance of the proposed method was compared with previous work, with the identification rate of the proposed method offering the best accuracy of 95.8%.