Files in this item



application/pdfZhen_Li.pdf (11MB)
(no description provided)PDF


Title:Generative and discriminative models for person verification and efficient search
Author(s):Li, Zhen
Director of Research:Huang, Thomas S.
Doctoral Committee Chair(s):Huang, Thomas S.
Doctoral Committee Member(s):Hasegawa-Johnson, Mark A.; Liang, Feng; Liang, Zhi-Pei
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Subject(s):Person Verification
Efficient Person Search
Abstract:This dissertation studies the person verification problem in modern surveillance and video retrieval systems. The problem is to identify whether a pair of face or human body images is about the same person, even if the person is not seen before. Traditional methods either model the intrapersonal and extrapersonal variations with probabilistic distributions, or look for a distance (or similarity) measure between images (e.g., by metric learning algorithms), and make decisions based on a fixed threshold. We show that the resulting decisions, depending merely on pairwise image differences, are nevertheless insufficient and sub-optimal for the verification problem. In this dissertation, we study both generative and discriminative models for person verification. Both methods consider a joint model of two images in a pair, and provide a decision function of second-order form that generalizes from previous approaches. We also generalize our model to a multi-setting scenario, where environment mismatch, a major challenge in cross-setting person verification, is handled. We evaluate our algorithms on face verification and human body verification problems on a number benchmark datasets, such as Multi-PIE, LFW, CIGIT-AIS, VIPer, VIPeR, and CAVIAR4REID. Our methods outperform not only the classical Bayesian Face Recognition approach, metric learning algorithms (LMNN, ITML, etc.), but also the state-of-the-art in the computer vision community. This dissertation also considers efficient person search, a potential application of person verification in surveillance systems. To this end, we propose a general learning-to-search framework for efficient similarity search in high dimensions. Experimental results show that our approach significantly outperforms the state-of-the-art learning-to-hash methods (such as spectral hashing), as well as state-of-the-art high-dimensional search algorithms (such as LSH and k-means trees).
Issue Date:2013-05-24
Rights Information:Copyright 2013 Zhen Li
Date Available in IDEALS:2013-05-24
Date Deposited:2013-05

This item appears in the following Collection(s)

Item Statistics