Machine Learning for Multimedia Content Analysis

by Gong, Yihong; Xu, Wei

• ISBN13: 9780387699387
• ISBN10: 0387699384
• Format: Hardcover
• Copyright: 2007-10-15
• Publisher: Springer-Verlag New York Inc

Summary

Challenges in complexity and variability of multimedia data have led to revolutions in machine learning techniques. Multimedia data, such as digital images, audio streams and motion video programs, exhibit richer structures than simple, isolated data items. A number of pixels in a digital image collectively conveys certain visual content to viewers. A TV video program consists of both audio and image streams that unfold the underlying story. To recognize the visual content of a digital image, or to understand the underlying story of a video program, we may need to label sets of pixels or groups of image and audio frames jointly. Certain techniques exploit spatial, temporal structures, and model the correlations among different elements of the target problems within the machine learning field. Machine Learning Techniques for Multimedia introduces machine learning techniques that are particularly powerful and effective for modeling multimedia data and common tasks of multimedia content analysis. This book systematically covers these techniques in an intuitive fashion and demonstrates their applications through case studies.

Table of Contents

Introduction	p. 1
Basic Statistical Learning Problems	p. 2
Categorizations of Machine Learning Techniques	p. 4
Unsupervised vs. Supervised	p. 4
Generative Models vs. Discriminative Models	p. 4
Models for Simple Data vs. Models for Complex Data	p. 6
Model Identification vs. Model Prediction	p. 7
Multimedia Content Analysis	p. 8
Unsupervised Learning
Dimension Reduction	p. 15
Objectives	p. 15
Singular Value Decomposition	p. 16
Independent Component Analysis	p. 20
Preprocessing	p. 23
Why Gaussian is Forbidden	p. 24
Dimension Reduction by Locally Linear Embedding	p. 26
Case Study	p. 30
Problems	p. 34
Data Clustering Techniques	p. 37
Introduction	p. 37
Spectral Clustering	p. 39
Problem Formulation and Criterion Functions	p. 39
Solution Computation	p. 42
Example	p. 46
Discussions	p. 50
Data Clustering by Non-Negative Matrix Factorization	p. 51
Single Linear NMF Model	p. 52
Bilinear NMF Model	p. 55
Spectral vs. NMF	p. 59
Case Study: Document Clustering Using Spectral and NMF Clustering Techniques	p. 61
Document Clustering Basics	p. 62
Document Corpora	p. 64
Evaluation Metrics	p. 64
Performance Evaluations and Comparisons	p. 65
Generative Graphical Models
Introduction of Graphical Models	p. 73
Directed Graphical Model	p. 74
Undirected Graphical Model	p. 77
Generative vs. Discriminative	p. 79
Content of Part II	p. 80
Markov Chains and Monte Carlo Simulation	p. 81
Discrete-Time Markov Chain	p. 81
Canonical Representation	p. 84
Definitions and Terminologies	p. 88
Stationary Distribution	p. 91
Long Run Behavior and Convergence Rate	p. 94
Markov Chain Monte Carlo Simulation	p. 100
Objectives and Applications	p. 100
Rejection Sampling	p. 101
Markov Chain Monte Carlo	p. 104
Rejection Sampling vs. MCMC	p. 110
Problems	p. 112
Markov Random Fields and Gibbs Sampling	p. 115
Markov Random Fields	p. 115
Gibbs Distributions	p. 117
Gibbs - Markov Equivalence	p. 120
Gibbs Sampling	p. 123
Simulated Annealing	p. 126
Case Study: Video Foreground Object Segmentation by MRF	p. 133
Objective	p. 134
Related Works	p. 134
Method Outline	p. 135
Overview of Sparse Motion Layer Computation	p. 136
Dense Motion Layer Computation Using MRF	p. 138
Bayesian Inference	p. 140
Solution Computation by Gibbs Sampling	p. 141
Experimental Results	p. 143
Problems	p. 146
Hidden Markov Models	p. 149
Markov Chains vs. Hidden Markov Models	p. 149
Three Basic Problems for HMMs	p. 153
Solution to Likelihood Computation	p. 154
Solution to Finding Likeliest State Sequence	p. 158
Solution to HMM Training	p. 160
Expectation-Maximization Algorithm and its Variances	p. 162
Expectation-Maximization Algorithm	p. 162
Baum-Welch Algorithm	p. 164
Case Study: Baseball Highlight Detection Using HMMs	p. 167
Objective	p. 167
Overview	p. 167
Camera Shot Classification	p. 169
Feature Extraction	p. 172
Highlight Detection	p. 173
Experimental Evaluation	p. 174
Problems	p. 175
Inference and Learning for General Graphical Models	p. 179
Introduction	p. 179
Sum-product algorithm	p. 182
Max-product algorithm	p. 188
Approximate inference	p. 189
Learning	p. 191
Problems	p. 196
Discriminative Graphical Models
Maximum Entropy Model and Conditional Random Field	p. 201
Overview of Maximum Entropy Model	p. 202
Maximum Entropy Framework	p. 204
Feature Function	p. 204
Maximum Entropy Model Construction	p. 205
Parameter Computation	p. 208
Comparison to Generative Models	p. 210
Relation to Conditional Random Field	p. 213
Feature Selection	p. 215
Case Study: Baseball Highlight Detection Using Maximum Entropy Model	p. 217
System Overview	p. 218
Highlight Detection Based on Maximum Entropy Model	p. 220
Multimedia Feature Extraction	p. 222
Multimedia Feature Vector Construction	p. 226
Experiments	p. 227
Problems	p. 232
Max-Margin Classifications	p. 235
Support Vector Machines (SVMs)	p. 236
Loss Function and Risk	p. 237
Structural Risk Minimization	p. 237
Support Vector Machines	p. 239
Theoretical Justification	p. 243
SVM Dual	p. 244
Kernel Trick	p. 245
SVM Training	p. 248
Further Discussions	p. 255
Maximum Margin Markov Networks	p. 257
Primal and Dual Problems	p. 257
Factorizing Dual Problem	p. 259
General Graphs and Learning Algorithm	p. 262
Max-Margin Networks vs. Other Graphical Models	p. 262
Problems	p. 264
Appendix	p. 267
References	p. 269
Index	p. 275
Table of Contents provided by Ingram. All Rights Reserved.