Wavelets and Gaussian mixture model approach for gender classification using fingerprints

Gender classification is the most challenging task in forensic investigation. In this paper, a new approach to estimate gender by multiresolutional analysis of fingerprints is proposed. Discrete Wavelet Transform (DWT) is used to analyze the fingerprints in the frequency domain. The classification task is modeled by gaussian mixtures. DWT coefficients are used as features and only dominant features selected by ranking are fed into GMM for classification. This system carried out with the database of 180 persons in which 80 are females and 100 are males. The results show that the proposed system achieves 92.67% at 3rd level DWT decomposition with 16 gaussian densities.