Face and Gender Recognition System Based on Convolutional Neural networks

Abstract

In the existing research, face features and gender attributes are separated, resulting in face recognition errors and gender recognition errors in complex backgrounds. In this work, we propose the Face and Gender Recognition System that uses convolutional neural networks (CNN). The system consists of two components: one is face recognition module and two is gender recognition module. Both face recognition module and gender recognition module use pre-trained CNN to extract face and gender features in the image. Specifically, in the face recognition module, we use the public datasets Labeled Faces in the Wild (LFW), YouTube Face (YTF) and VGGFace2 to train CNN, which improves the precision. In the gender recognition module, we use the public dataset Adience to train CNN and improve the best recognition accuracy from 91.80% to 93.22%.