Gender Classification Based on Fingerprint Using Wavelet and Multilayer Perceptron

Abstract

Fingerprint-based gender classification is beneficial for speeding up the fingerprint identification of criminals, accident victims, and natural disaster victims that are difficult to be recognized based on their physical characteristics. The biggest obstacle to digitally classifying fingerprints is the image's poor quality. Some methods have been developed to improve image quality through various preprocessing, such as noise removal, background segmentation, thinning, and binarization. However, as these processes increase the classification time, some methods have been developed to classify fingerprints without preprocessing. One of them that has shown excellent success is CNN (Convolutional Neural Network). The method does not require preprocessing, but the computation time is very long and requires large amounts of training data. This study proposed a new method that did not need any preprocessing by using wavelet decomposition combined with the max-pooling process to generate features. Firstly, the fingerprint image was decomposed with a Haar wavelet of 4 levels, and each level was followed by a max-pooling process with a 2´2 filter. After that, the resulting feature was used as training data for the Multilayer Perceptron (MLP) network. In this study, the training data was a dataset from NIST (National Institute of Standart and Technology), with 750 fingerprints consisting of male and female fingerprints, each as many as 375. The method could produce a total accuracy of 80.1%.