3D Morphable Ear Model: A Complete Pipeline from Ear Segmentation to Statistical Modeling

Abstract

The shape of human ear contains crucial information that can be used for biometric identification. Analysis of the ear shape can be improved by using a statistical shape model known as 3D Morphable Ear Model (3DMEM). In this work, we propose a complete pipeline to create the 3DMEM by following a three-step procedure. First, a large ear database is created by segmenting ears from 3D profile faces using a deep convolutional neural network. Next, dense correspondence between 3D ears is established using Generalized Procrustes Analysis (GPA). Finally, the 3DMEM is constructed using Principal Component Analysis (PCA). Our results show that 3DMEM can generalize well on unseen 3D ear data.