Learning Discriminating Features for Gender Recognition of Real World Faces

Abstract

The automatic gender recognition of faces has many applications, for example surveillance, targeted advertisement and human computer interaction, etc. Humans have the ability to accurately determine the gender from faces, however, for a machine, it is a difficult task. Many studies have targeted this problem, but most of these studies have used images taken under constrained conditions. In Real-world systems have to process images with wide variations in lighting and pose that makes the classification task very challenging. We have analyzed the gender classification of real world faces. Faces from images are detected, aligned and represented using local binary pattern histograms. Adaptive boosting selects the discriminating features and boosted LBP features are used to train a support vector machine that provides a recognition rate of 95.5%.