Robust gender classification on unconstrained face images

Abstract

Robust and automatic gender classification on a single face image is one of the fundamental artificial intelligence tasks. And it has become relevant to an increasing amount of applications alongside the rise of social media. In this paper we consider the problem of robust gender classification on unconstrained face images based on weakly calibration and deep neural network methods. While there has been a lot of significant researches on this problem, the proposed method has distinct advantages compared with other approaches. To facilitate tolerance of the pose variations of the face image caused by different shooting angles, an efficient face detection and calibration method is proposed to preprocess unconstrained face images. Furthermore, we extract deep face representations using deep convolutional neural networks (CNN) which could handle all unconstrained face images efficiently. We evaluate our method on the real-life unconstrained faces database-the Labeled Faces in the Wild (LFW) and it outperforms the previous best results reported in the literature.