Date of Award

2010

Publication Type

Doctoral Thesis

Degree Name

Ph.D.

Department

Electrical and Computer Engineering

Keywords

Engineering, Electronics and Electrical.

Supervisor

Wu, Jonathan (Electrical and Computer Engineering)

Rights

info:eu-repo/semantics/openAccess

Abstract

This dissertation is dedicated to the development of new algorithms for biometric applications based on multiresolution analysis tools. Biometric is a unique, measurable characteristic of a human being that can be used to automatically recognize an individual or verify an individual's identity. Biometrics can measure physiological, behavioral, physical and chemical characteristics of an individual. Physiological characteristics are based on measurements derived from direct measurement of a part of human body, such as, face, fingerprint, iris, retina etc. We focussed our investigations to fingerprint and face recognition since these two biometric modalities are used in conjunction to obtain reliable identification by various border security and law enforcement agencies. We developed an efficient and robust human face recognition algorithm for potential law enforcement applications. A generic fingerprint compression algorithm based on state of the art multiresolution analysis tool to speed up data archiving and recognition was also proposed. Finally, we put forth a new fingerprint matching algorithm by generating an efficient set of fingerprint features to minimize false matches and improve identification accuracy. Face recognition algorithms were proposed based on curvelet transform using kernel based principal component analysis and bidirectional two-dimensional principal component analysis and numerous experiments were performed using popular human face databases. Significant improvements in recognition accuracy were achieved and the proposed methods drastically outperformed conventional face recognition systems that employed linear one-dimensional principal component analysis. Compression schemes based on wave atoms decomposition were proposed and major improvements in peak signal to noise ratio were obtained in comparison to Federal Bureau of Investigation's wavelet scalar quantization scheme. Improved performance was more pronounced and distinct at higher compression ratios. Finally, a fingerprint matching algorithm based on wave atoms decomposition, bidirectional two dimensional principal component analysis and extreme learning machine was proposed and noteworthy improvements in accuracy were realized.

Share

COinS