Multi-model Ear Database for Biometric Applications

摘要

We present the 3DEarDB, a multi-model ear database, characterized by different types of ear representation, either 2D or 3D, depending on the acquisition device used. The main objective is to provide the biometrics community with a unified tool for testing and comparing of classification algorithms not only on 2D intensity and/or depth images, or videos, but also on detailed 3D mesh models of human ears. The 3DEarDB features accurate 3D mesh models of right ear captured from more than 100 subjects, with a resolution of 1 mm and an accuracy of 0.05 mm, collected via the VIUscan 3D laser scanner, available at the Smart Lab of IICT-BAS, in the AComIn project frames. Two more ear acquisition modalities are also included: 3D Kinect ear depth maps and 2D high-definition video clips, associated to the basic mesh models. To extend 3DEarDB compatibilities with known methods for 2D/3D ear detection and/or recognition, we provide two more ear model types. Namely, a set of 2D ear intensity projections (of different orientations and/or lightening directions), and a set of 2D depth map projections can be generated by demand from the basic 3D ear models. Finally, we report about preliminary experiments conducted by means of Extended Gaussian Image approach that confirm the consistency of the proposed 3D-Ear-Data-Base.