Gender classification using automatically detected and aligned 3D ear range data

Gender classification received attention due to its use in many applications. In this paper, the potential of using the 3D shape of the ear for gender recognition is established. We demonstrate the first attempt for gender classification from 3D ear data and evaluate different algorithms using automatically detected and aligned ears. Experiments were conducted on the University of Notre Dame (UND) database collections F and J2 which contain images with large occlusion and pose variations. It is observed that the use of Histogram of Indexed Shapes (HIS) feature along with Support Vector Machine (SVM) yields an average classification accuracy of 92.94%, which is superior to the state-of-the-art for gender classification from 2D ear images, and shows that the 3D shape of the ear comprises rich gender information.