University of Notre Dame
Browse

File(s) under permanent embargo

Ear Biometrics in Human Identification

thesis
posted on 2006-07-07, 00:00 authored by Ping Yan
Biometrics are physical or behavioral characteristics that can be used for human identification. Security plays an increasingly important role in our daily life, and biometric technologies are becoming the solution to highly secure recognition and verification of identity. In this dissertation, we propose the ear as a biometric and investigate its potential with both 2D and 3D data. Our work is the largest experimental investigation of ear biometrics to date. The data set used for our experiments contains 415 persons, each with images acquired on at least two different dates. Approaches considered include a PCA ('eigen-ear') approach with 2D intensity images and range images, a Hausdorff matching of edges from range images, and an ICP-based approach to matching the 3D data. Our experimental results demonstrate that the ICP-based approach outperforms the other approaches at a statistically significant level. Furthermore, we develop the first fully automated biometric system using 3D ear shape. It is a complete automatic system starting with a raw 3D image, through automated segmentation of the ear, and 3D shape matching for recognition. The automatic system achieves a rank-one recognition rate of 97.6% on a 415-subject dataset. Our algorithm also shows good scalability of recognition rate with size of dataset. The results suggest a strong potential for 3D ear shape as a biometric. In a biometrics scenario, gallery images are enrolled into the database ahead of the matching step, which provides the opportunity to build related information before the probe comes into the system. We present a novel approach, called 'Pre-computed Voxel Closest Neighbors', to reduce the computational time for shape matching in a biometrics context. The approach shifts the heavy computation burden to the enrollment stage, which can be done offline. Experiments in 3D ear and face biometrics demonstrate the effectiveness of the approach. In addition, ear symmetry and partial ear shape experiments are investigated. The results indicate that most people's left and right ears are approximately bilaterally symmetric. However, some people have ears with recognizably different shapes. Experimental results with partial ear shape suggest that minor hair covering does not affect the performance substantially, but large hair covering will certainly reduce the recognition rate. This suggests that even in circumstances where the complete ear shape cannot be captured, partial shape has potential for recognition. This lends support for using ear shape as a biometric. Our experiments use the biometric database collected at the University of Notre Dame. This data set is available to other research groups.

History

Date Modified

2017-06-05

Defense Date

2006-05-19

Research Director(s)

Steven R. Schmid

Committee Members

Amitabh Chaudhary Kevin W. Bowyer Patrick J. Flynn Gregory Madey

Degree

  • Doctor of Philosophy

Degree Level

  • Doctoral Dissertation

Language

  • English

Alternate Identifier

etd-07072006-101848

Publisher

University of Notre Dame

Program Name

  • Computer Science and Engineering (CSE)

Usage metrics

    Dissertations

    Categories

    No categories selected

    Exports

    RefWorks
    BibTeX
    Ref. manager
    Endnote
    DataCite
    NLM
    DC