PESNET: A CNN FRAMEWORK TO PREDICT THE GENDER OF HUMANS USING BARE FOOTPRINTS

Abstract

In forensic science, one of the common applications of personal identification occurs in cases involving unidentified human remains. This work proposes PesNet, a gender prediction framework from bare footprints using a novel convolutional neural network (CNN). In this study, a dataset of 1,000 human barefoot impressions, evenly split between male and female sample impressions, collected from the different administrative division of Haryana, India, was exploited to visualize and analyze the shape and sizes by utilizing deep learning approaches. PesNet architecture in this article uses higher order preprocessing methods such as Gaussian filtering and grayscale conversion to improve image quality and minimizes environmental noise. PesNet, deployed with a pre-trained model, prevailed over conventional methods and established models to achieve a stunning gender classification accuracy of 96% are the findings of our study. This research not only highlights the potential of deep learning in forensic applications but also addresses challenges such as class imbalance and the need for reliable identification techniques in criminal investigations.