A fingerprint-based hybrid gender classification system using genetic algorithm

Abstract

In this article, a multilevel hybrid approach which gives better accuracy for gender classification is presented. The first level uses discrete wavelet transform DWT, singular value decomposition SVD and principal component analysis PCA techniques to derive three independent sets of feature vectors for simultaneous gender classification by three independent neural networks. Coefficients similar to cdf9/7 DWT lifting coefficients and optimum values of initial seeds for the classifiers are evolved using genetic algorithm GA, for obtaining better feature vectors. Use of lifting coefficients causes faster evolution. In the second stage, the output is derived by decision formulated based on the outputs of the individual classifiers. With a database consisting left thumb impressions of 100 males and 100 females, an overall success rate of 93.94% and an average improvement of 5.24% accuracy over the existing classifiers is achieved. Use of feature vectors having lesser number of elements enhances the speed of operation of the classifier as well.