Improved 3D ear reconstruction based on 3D EMM

Ear has been proven to be a good candidate for non-contact recognition. In order to acquire ear´s 3D information as well as reserve its non-contact advantage, 3D reconstruction using 2D images can be a promising way. However ear is a small object with abundant structure information, which makes the 3D shape estimating a challenge problem. An improved sophisticated 3D ear reconstruction method based on 3D ear morphable model is demonstrated. We propose a novel ear contour feature points extraction method based on the automatically detection of ear contour which combines both intensity and depth image. Abundant experiment results show that more realistic 3D ear sample with smooth contour can be generated using the automatically detected feature points, not to mention it can greatly reduce the workload. After statistical model training and model fitting based on sparse points, the reconstruction accuracy is discussed. This is the first 3D ear reconstruction work which demonstrates the reconstruction accuracy quantitatively. Abundant experiment results show the efficiency of our proposed method.