Efficient 3D ear reconstruction by employing the 3D ear morphable model

Based on studies on the existing three-dimensional (3D) ear reconstruction method and morphable model theory, considering the unique structure of human ears, a novel 3D ear reconstruction method was proposed, which is based on the 3D ear morphable model. First, ear feature points on the contour were located by the perpendicular bisector method. Then, the dense match of 3D ear samples based on physiological characteristics was realized by a hierarchical triangle mesh method. The generalized Procrustes analysis was performed, and the ear shape alignment was achieved accurately and automatically in the 3D space. Finally, the 3D ear morphable model was trained by using 3D ear data from the University of Notre Dame (UND) 3D ear dataset. This method can reconstruct the corresponding dense 3D model using only one 2D ear image. Extensive experimental results on UND collection J2 dataset and USTB ear image database for reconstruction show that the Drooosed method is effective and advantageous.