Ouyang, Hua. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2007. / Includes bibliographical references (leaves 81-89). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Biometric Authentication --- p.1 / Chapter 1.2 --- Multimodal Biometric Authentication --- p.2 / Chapter 1.2.1 --- Combination of Different Biometric Traits --- p.3 / Chapter 1.2.2 --- Multimodal Fusion --- p.5 / Chapter 1.3 --- Audio-Visual Bi-modal Authentication --- p.6 / Chapter 1.4 --- Focus of This Research --- p.7 / Chapter 1.5 --- Organization of This Thesis --- p.8 / Chapter 2 --- Audio-Visual Bi-modal Authentication --- p.10 / Chapter 2.1 --- Audio-visual Authentication System --- p.10 / Chapter 2.1.1 --- Why Audio and Mouth? --- p.10 / Chapter 2.1.2 --- System Overview --- p.11 / Chapter 2.2 --- XM2VTS Database --- p.12 / Chapter 2.3 --- Visual Feature Extraction --- p.14 / Chapter 2.3.1 --- Locating the Mouth --- p.14 / Chapter 2.3.2 --- Averaged Mouth Images --- p.17 / Chapter 2.3.3 --- Averaged Optical Flow Images --- p.21 / Chapter 2.4 --- Audio Features --- p.23 / Chapter 2.5 --- Video Stream Classification --- p.23 / Chapter 2.6 --- Audio Stream Classification --- p.25 / Chapter 2.7 --- Simple Fusion --- p.26 / Chapter 3 --- Weighted Sum Rules for Multi-modal Fusion --- p.27 / Chapter 3.1 --- Measurement-Level Fusion --- p.27 / Chapter 3.2 --- Product Rule and Sum Rule --- p.28 / Chapter 3.2.1 --- Product Rule --- p.28 / Chapter 3.2.2 --- Naive Sum Rule (NS) --- p.29 / Chapter 3.2.3 --- Linear Weighted Sum Rule (WS) --- p.30 / Chapter 3.3 --- Optimal Weights Selection for WS --- p.31 / Chapter 3.3.1 --- Independent Case --- p.31 / Chapter 3.3.2 --- Identical Case --- p.33 / Chapter 3.4 --- Confidence Measure Based Fusion Weights --- p.35 / Chapter 4 --- Regularized k-Nearest Neighbor Classifier --- p.39 / Chapter 4.1 --- Motivations --- p.39 / Chapter 4.1.1 --- Conventional k-NN Classifier --- p.39 / Chapter 4.1.2 --- Bayesian Formulation of kNN --- p.40 / Chapter 4.1.3 --- Pitfalls and Drawbacks of kNN Classifiers --- p.41 / Chapter 4.1.4 --- Metric Learning Methods --- p.43 / Chapter 4.2 --- Regularized k-Nearest Neighbor Classifier --- p.46 / Chapter 4.2.1 --- Metric or Not Metric? --- p.46 / Chapter 4.2.2 --- Proposed Classifier: RkNN --- p.47 / Chapter 4.2.3 --- Hyperkernels and Hyper-RKHS --- p.49 / Chapter 4.2.4 --- Convex Optimization of RkNN --- p.52 / Chapter 4.2.5 --- Hyper kernel Construction --- p.53 / Chapter 4.2.6 --- Speeding up RkNN --- p.56 / Chapter 4.3 --- Experimental Evaluation --- p.57 / Chapter 4.3.1 --- Synthetic Data Sets --- p.57 / Chapter 4.3.2 --- Benchmark Data Sets --- p.64 / Chapter 5 --- Audio-Visual Authentication Experiments --- p.68 / Chapter 5.1 --- Effectiveness of Visual Features --- p.68 / Chapter 5.2 --- Performance of Simple Sum Rule --- p.71 / Chapter 5.3 --- Performances of Individual Modalities --- p.73 / Chapter 5.4 --- Identification Tasks Using Confidence-based Weighted Sum Rule --- p.74 / Chapter 5.4.1 --- Effectiveness of WS_M_C Rule --- p.75 / Chapter 5.4.2 --- WS_M_C v.s. WS_M --- p.76 / Chapter 5.5 --- Speaker Identification Using RkNN --- p.77 / Chapter 6 --- Conclusions and Future Work --- p.78 / Chapter 6.1 --- Conclusions --- p.78 / Chapter 6.2 --- Important Follow-up Works --- p.80 / Bibliography --- p.81 / Chapter A --- Proof of Proposition 3.1 --- p.90 / Chapter B --- Proof of Proposition 3.2 --- p.93
| Identifer | oai:union.ndltd.org:cuhk.edu.hk/oai:cuhk-dr:cuhk_326068 |
| Date | January 2007 |
| Contributors | Ouyang, Hua., Chinese University of Hong Kong Graduate School. Division of Electronic Engineering. |
| Source Sets | The Chinese University of Hong Kong |
| Language | English, Chinese |
| Detected Language | English |
| Type | Text, bibliography |
| Format | print, xvi, 95 leaves : ill. ; 30 cm. |
| Rights | Use of this resource is governed by the terms and conditions of the Creative Commons “Attribution-NonCommercial-NoDerivatives 4.0 International” License (http://creativecommons.org/licenses/by-nc-nd/4.0/) |
Page generated in 0.0027 seconds