NP-PHOG: An Automated Gender Classification Model based on Nested Patch-based Prymadial Histogram-Oriented Gradients Feature Extraction using Shadow Images

Abstract

Gender classification is a fundamental area of research in machine learning, and numerous types of data, such as gaits, faces, and speeches, have been utilized for gender classification. In this research, we introduce a novel data type, namely shadow images, for detecting gender. We collected a shadow image dataset comprising two classes, namely (1) female and (2) male. To propose an automated shadow gender classification model, we developed a pyramidal histogram-oriented gradient (PHOG) based model. Our model consists of three primary phases, including (i) feature extraction using nested patches and PHOG, (ii) neighborhood component analysis (NCA) based feature selection, and (iii) classification with support vector machine (SVM) classifier. Therefore, we have named our model nested patches (NP) based PHOG – NP-PHOG–. The proposed NP-PHOG model was applied to the collected shadow image dataset, and it achieved a classification accuracy of 99.69%. This result provides strong evidence that machine learning models can accurately detect gender using shadows, and new image forensic tools can be developed using this approach.