声明
摘要
ABSTRACT
Table of Contents
List of Figures
List of Tables
Chapter 1 Introduction
1.1 Introducfion
1.2 Motivations
1.3 Contributions
1.4 Dissertation Organization
Chapter 2 Literature Review
2.1 Introduction
2.2 Appearance-based Person Re-Id
2.2.1 Feature Representation
2.2.2 Feature Description
2.3 Metric Learning Methods
2.3.1 Large Margin Nearest Neighbor (LMNN)
2.3.3 Probabilistic Relative Distance Comparison (PRDC)
2.3.5 Cross-view Quadratic Discriminant Analysis (XQDA)
2.3.6 Symmetry-Driven Accumulation of Local Features (SDALF)
2.4 Post-Rank Optimization and Prioritization Methods
2.4.1 Soft Biometrics Attribute-based Re-ranking
2.4.2 Post-rank Optimization (POP)
2.4.3 Bi-directional Re-ranking
2.4.4 Saliency Re-ranking
2.4.5 Discriminative Content and Context Analysis (DCIA)
2.5 Evaluation Metrics
2.6 Discussions
Chapter 3 Pre-rank Prioritization
3.2 System Overview
3.2.1 Person Tracking and Detection
3.2.2 Color-based Pre-ranking
3.2.3 Signature Generation and Training
3.3 Results and Discussions
3.3.1 Datasets
3.3.2 Feature Extraction and Evaluation Protocols
3.3.3 Comparison with Other Person Re-Id Methods
3.4 Summary
Chapter 4 Post-rank Optimization via Hypergraphs
4.1 Introduction:Post-rank Optimization and Prioritization
4.2 Hypergraph-based Post-rank Optimization
4.2.1 Motivations of Using Hypergraph
4.2.2 Basic Notations Used in Hypergraph
4.3 System Overview
4.3.1 The Rank List Refinement
4.3.2 Hypergraph Learning for Re-ranking
4.3.3 Weight Learning of Hyperedges
4.4 Experiments and Results
4.4.1 Datasets
4.4.2 Feature Extraction and Evaluation Protocols
4.4.3 Evaluation with State-of-the-art Post-ranking Approaches
4.4.4 Evaluation with State-of-the-art Ranking Approaches
4.5 Summary
Chapter 5 Multi-feature Fusion Based Rank Optimization
5.1 Introduction:Multi-feature Fusion
5.2 System Overview
5.2.1 Multi-feature Selection and Fusion
5.2.2 Low Dimensional Embedding
5.2.3 Image Tree-based Re-ranking
5.3 Experiments and Results
5.3.1 Datasets
5.3.2 Evaluation Setting
5.3.3 Comparison with Other Methods
5.3.4 Comparison with Single Features
5.4 Summary
Chapter 6 POP:System Design and Performance Evaluation Considerations
6.1.1 POP Methods:Key to Good Results/Prioritization
6.1.2 Feature/Descriptor Level Challenges
6.1.3 Baseline Method Selection
6.1.4 Benchmark Datasets
6.1.5 Experimental Setup and Evaluation Protocols
6.2 Summary
Chapter 7 Discussions
7.1 Concluding Remarks and Summary of Contributions
7.2 Future Work
Appendix
References
ACKOWLEDGEMENT
List of Publications