Age, Gender and Race Estimation from Unconstrained Face Images

Automatic estimation of demographic attributes (e.g., age, gender, and race) from a face image is a topic of growing interest with many potential applications. Most prior work on this topic has used face images acquired under constrained and cooperative scenarios. This paper addresses the more challenging problem of automatic age, gender, and race estimation from real-life face images (face images in the wild) acquired in unconstrained conditions. Given an input face image, we first normalize it by performing pose and photometric corrections. Biologically inspired features (BIF) are then extracted from the normalized face image, including both the central face region and the surrounding context region. Given this representation, three different Support Vector Machines (SVM) are used to predict the age group (or exact age), gender, and race of a subject. Experimental results on two large public-domain unconstrained face databases (Images of Groups and LFW) show that the proposed approach significantly outperforms the stateof-the-art methods. Our results also highlight that extraction of demographic attributes from face images in the wild is a difficult problem.