Gender Classification Based on Fingerprint Using Sobel Filter and Artificial Neural Network

Determining gender based on fingerprints is essential in many critical circumstances. Using fingerprints as biometrics is simple and reliable. However, a complex procedure is needed to choose reliable features and appropriate methods to get high-accuracy results. This study aims to estimate gender based on fingerprints using simple features to reduce the need for lengthy computation in subsequent processes. We select a small region of interest (RDI) of the fingerprint to create features. The RDI size is 80% of the image size at the centre. For this RDI selection, we created a custom function with MATLAB. We base our features on Sobel filters and arrange them into histograms. Furthermore, we classify the resulting histogram using a single-layer Artificial Neural Network (ANN). We tested the model with the Sokoto Coventry Fingerprint Dataset datasets, which comprise 1000 male and 1000 female fingerprints. Despite its simplicity, our model has demonstrated a validation accuracy of up to 68.2% on the dataset, which is lower than the results of more complex Convolutional Neural Networks (CNN) models. The model accurately predicted males at 71.5% and females at 65.2%.