An efficient technique for ear detection in 3D: Invariant to rotation and scale

This paper proposes an efficient technique to detect ear from 3D profile face range images and is invariant to rotation and scale. It makes use of graph connected components constructed using the edges of the depth map image of the range data for ear detection. It can detect left and right ears at the same time without imposing any additional cost or specific training. The technique has been tested on University of Notre Dame 3D profile face database, Collection J2 (UND-J2) having range images with scale and pose variations. Results are also compared with the well known 3D ear detection techniques to show its superiority.