Gender Recognition from Standard Images

Abstract

Human gender plays an imperative role as a social construct and an essential form of an individual’s personality. This is highly reflected in social communication, forensic science, surveillance, and targeted marketing. The task has always been practiced in conjunction with face recognition, as it yields high performance compared to other features, e.g., gait and sound. This is attributed to its richness with diverse useful clues for gender inspection. However, facial features were always used separately without any integration between them, and a definitive solution has not yet been found yet. We propose a robust feature fusion method that integrates various facial features, in addition to a set of new features. Furthermore, Haar-like features are used to extract the essential features, while the new features used are, i.e., beard, moustache, cheeks, forehead, and face. The scientific novelty of the research is the robust fusion of features to achieve high recognition accuracy. The proposed fusion method is a priority-based method that arranges classifiers’ responses from the strongest to weakest according to their importance to yield a final decision. When the proposed technique compares the recent and standard benchmarks, it achieves high recognition accuracy. The experiments on various standard and challenging datasets widely adopted in the scientific community, namely LFW, DataHub, NIST, and Caltech WebFaces, demonstrated a robust performance of 99.1% accuracy.