Real-Time Gender Recognition From Face Images Using Deep Convolutional Neural Network

BUEMI, M. E; I. OROZCO; JACOBO, J.; IGLESIAS F.

Valparaiso

Conferencia; Latin American Conference on Networked Electronic Media; 2017

, Universidad Técnica Federico Santa María (UTFSM), Valparaíso, Chile

Gender recognition is a topic of interest in computer vision dueto its applications such as surveillance in public places, directedadvertising, among others. The good results obtained usingdeep convolutional neural networks in vision tasks make theman attractive tool to improve the capacities of gender recogni-tion systems. In this work we propose a deep convolutionalnetwork architecture to classify as male or female person thecandidate regions previously detected using Haar features em-bedded in an AdaBoost scheme [14]. The data used for trainingand testing come from the Labeled Faces in the Wild [7] andGallagher?s dataset [6]. We have evaluated the classificationresults on the proposed architecture and have obtained an av-erage of ∼ 95.42% and ∼ 91.48% accuracy for the trainingset and for the test set, respectively, that are competitive withthose mentioned in the bibliography. We have also carried outa real-time evaluation of the system using a web camera.