Sebe / Cohen / Garg Machine Learning in Computer Vision

1. Auflage 2005
ISBN: 978-1-4020-3275-2
Verlag: Springer Netherlands
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, Band 29, 253 Seiten

Reihe: Computational Imaging and Vision

The goal of this book is to address the use of several important machine learning techniques into computer vision applications. An innovative combination of computer vision and machine learning techniques has the promise of advancing the field of computer vision, which contributes to better understanding of complex real-world applications.
<
The effective usage of machine learning technology in real-world computer vision problems requires understanding the domain of application, abstraction of a learning problem from a given computer vision task, and the selection of appropriate representations for the learnable (input) and learned (internal) entities of the system. In this book, we address all these important aspects from a new perspective: that the key element in the current computer revolution is the use of machine learning to capture the variations in visual appearance, rather than having the designer of the model accomplish this. As a bonus, models learned from large datasets are likely to be more robust and more realistic than the brittle all-design models.

Autoren/Hrsg.

Sebe, Nicu

Cohen, Ira

Garg, Ashutosh

Huang, Thomas S.

Inhaltsverzeichnis

1;Contents;7
2;Foreword;13
3;Preface;15
4;1 INTRODUCTION;18
4.1;1. Research Issues on Learning in Computer Vision;19
4.2;2. Overview of the Book;23
4.3;3. Contributions;29
5;2 THEORY: PROBABILISTIC CLASSIFIERS;32
5.1;1. Introduction;32
5.2;2. Preliminaries and Notations;35
5.2.1;2.1 Maximum Likelihood Classification;35
5.2.2;2.2 Information Theory;36
5.2.3;2.3 Inequalities;37
5.3;3. Bayes Optimal Error and Entropy;37
5.4;4. Analysis of Classification Error of Estimated (Mismatched) Distribution;44
5.4.1;4.1 Hypothesis Testing Framework;45
5.4.2;4.2 Classification Framework;47
5.5;5. Density of Distributions;48
5.5.1;5.1 Distributional Density;50
5.5.2;5.2 Relating to Classification Error;54
5.6;6. Complex Probabilistic Models and Small Sample Effects;57
5.7;7. Summary;58
6;3 THEORY: GENERALIZATION BOUNDS;62
6.1;1. Introduction;62
6.2;2. Preliminaries;64
6.3;3. A Margin Distribution Based Bound;66
6.3.1;3.1 Proving the Margin Distribution Bound;66
6.4;4. Analysis;74
6.4.1;4.1 Comparison with Existing Bounds;76
6.5;5. Summary;81
7;4 THEORY: SEMI-SUPERVISED LEARNING;82
7.1;1. Introduction;82
7.2;2. Properties of Classification;84
7.3;3. Existing Literature;85
7.4;4. Semi-supervised Learning Using Maximum Likelihood Estimation;87
7.5;5. Asymptotic Properties of Maximum Likelihood Estimation with Labeled and Unlabeled Data;90
7.5.1;5.1 Model Is Correct;93
7.5.2;5.2 Model Is Incorrect;94
7.5.3;5.3 Examples: Unlabeled Data Degrading Performance with Discrete and Continuous Variables;97
7.5.4;5.4 Generating Examples: Performance Degradation with Univariate Distributions;100
7.5.5;5.5 Distribution of Asymptotic Classi.cation Error Bias;103
7.5.6;5.6 Short Summary;105
7.6;6. Learning with Finite Data;107
7.6.1;6.1 Experiments with Artificial Data;108
7.6.2;6.2 Can Unlabeled Data Help with Incorrect Models? Bias vs. Variance Effects and the Labeled-unlabeled Graphs;109
7.6.3;6.3 Detecting When Unlabeled Data Do Not Change the Estimates;114
7.6.4;6.4 Using Unlabeled Data to Detect Incorrect Modeling Assumptions;116
7.7;7. Concluding Remarks;117
8;5 ALGORITHM: MAXIMUM LIKELIHOOD MINIMUM ENTROPY HMM;120
8.1;1. Previous Work;120
8.2;2. Mutual Information, Bayes Optimal Error, Entropy, and Conditional Probability;122
8.3;3. Maximum Mutual Information HMMs;124
8.3.1;3.1 Discrete Maximum Mutual Information HMMs;125
8.3.2;3.2 Continuous Maximum Mutual Information HMMs;127
8.3.3;3.3 Unsupervised Case;128
8.4;4. Discussion;128
8.4.1;4.1 Convexity;128
8.4.2;4.2 Convergence;129
8.4.3;4.3 Maximum A-posteriori View of Maximum Mutual Information HMMs;129
8.5;5. Experimental Results;132
8.5.1;5.1 Synthetic Discrete Supervised Data;132
8.5.2;5.2 Speaker Detection;132
8.5.3;5.3 Protein Data;134
8.5.4;5.4 Real-time Emotion Data;134
8.6;6. Summary;134
9;6 ALGORITHM: MARGIN DISTRIBUTION OPTIMIZATION;136
9.1;1. Introduction;136
9.2;2. A Margin Distribution Based Bound;137
9.3;3. Existing Learning Algorithms;138
9.4;4. The Margin Distribution Optimization (MDO) Algorithm;142
9.4.1;4.1 Comparison with SVM and Boosting;143
9.4.2;4.2 Computational Issues;143
9.5;5. Experimental Evaluation;144
9.6;6. Conclusions;145
10;7 ALGORITHM: LEARNING THE STRUCTURE OF BAYESIAN NETWORK CLASSIFIERS;146
10.1;1. Introduction;146
10.2;2. Bayesian Network Classifiers;147
10.2.1;2.1 Naive Bayes Classifiers;149
10.2.2;2.2 Tree-Augmented Naive Bayes Classifiers;150
10.3;3. Switching between Models: Naive Bayes and TAN Classifiers;155
10.4;4. Learning the Structure of Bayesian Network Classifiers: Existing Approaches;157
10.4.1;4.1 Independence-based Methods;157
10.4.2;4.2 Likelihood and Bayesian Score-based Methods;159
10.5;5. Classification Driven Stochastic Structure Search;160
10.5.1;5.1 Stochastic Structure Search Algorithm;160
10.5.2;5.2 Adding VC Bound Factor to the Empirical Error Measure;162
10.6;6. Experiments;163
10.6.1;6.1 Results with Labeled Data;163
10.6.2;6.2 Results with Labeled and Unlabeled Data;164
10.7;7. Should Unlabeled Data Be Weighed Differently?;167
10.8;8. Active Learning;168
10.9;9. Concluding Remarks;170
11;8 APPLICATION: OFFICE ACTIVITY RECOGNITION;174
11.1;1. Context-Sensitive Systems;174
11.2;2. Towards Tractable and Robust Context Sensing;176
11.3;3. Layered Hidden Markov Models (LHMMs);177
11.3.1;3.1 Approaches;178
11.3.2;3.2 Decomposition per Temporal Granularity;179
11.4;4. Implementation of SEER;181
11.4.1;4.1 Feature Extraction and Selection in SEER;181
11.4.2;4.2 Architecture of SEER;182
11.4.3;4.3 Learning in SEER;183
11.4.4;4.4 Classification in SEER;183
11.5;5. Experiments;183
11.5.1;5.1 Discussion;186
11.6;6. Related Representations;187
11.7;7. Summary;189
12;9 APPLICATION: MULTIMODAL EVENT DETECTION;192
12.1;1. Fusion Models: A Review;193
12.2;2. A Hierarchical Fusion Model;194
12.2.1;2.1 Working of the Model;195
12.2.2;2.2 The Duration Dependent Input Output Markov Model;196
12.3;3. Experimental Setup, Features, and Results;199
12.4;4. Summary;200
13;10 APPLICATION: FACIAL EXPRESSION RECOGNITION;204
13.1;1. Introduction;204
13.2;2. Human Emotion Research;206
13.2.1;2.1 Affective Human-computer Interaction;206
13.2.2;2.2 Theories of Emotion;207
13.2.3;2.3 Facial Expression Recognition Studies;209
13.3;3. Facial Expression Recognition System;214
13.3.1;3.1 Face Tracking and Feature Extraction;214
13.3.2;3.2 Bayesian Network Classifiers: Learning the “Structure” of the Facial Features;217
13.4;4. Experimental Analysis;218
13.4.1;4.1 Experimental Results with Labeled Data;221
13.4.1.1;4.1.1 Person-dependent Tests;222
13.4.1.2;4.1.2 Person-independent Tests;223
13.4.2;4.2 Experiments with Labeled and Unlabeled Data;224
13.5;5. Discussion;225
14;11 APPLICATION: BAYESIAN NETWORK CLASSIFIERS FOR FACE DETECTION;228
14.1;1. Introduction;228
14.2;2. Related Work;230
14.3;3. Applying Bayesian Network Classifiers to Face Detection;234
14.4;4. Experiments;235
14.5;5. Discussion;239
15;References;242
16;Index;254

Leseproben

Chapter 11

APPLICATION: BAYESIAN NETWORK CLASSIFIERS FOR FACE DETECTION (p.211)

Images containing faces are essential to intelligent vision-based human computer interaction. To buld fully automated systems that analyze the information contained in face images, robust and ef.cient face detection algorithms are required. Among the face detection methods, the ones based on learning algorithms have attracted much attention recently and have demonstrated excellent results.

This chapter presents a discussion on semi-supervised learning of probabilistic mixture model classi.ers for face detection. Based on our complete theoretical analysis of semi-supervised learning using maximum likelihood presented in Chapter 4 we discuss the possibility of structure learning of Bayesian networks for face detection. We show that learning the structure of Bayesian networks classi.ers enables learning of good classi.ers for face detection with a small labeled set and a large unlabeled set.

1. Introduction
Many of the recent applications designed for human-computer intelligent interaction applications have used the human face as an input. Systems that perform face tracking for various applications, facial expression recognition and pose estimation of faces all rely on detection of human faces in the video frames [Pentland, 2000]. The rapidly expanding research in face processing is based on the premise that information about user’s identity, state, and intend can be extracted from images and that computers can react accordingly, e.g., by observing a person’s facial expression.

In the last years, face and facial expression recognition have attracted much attention despite the fact that they have been studied for more than 20 years by psychophysicists, neuroscientists, and engineers. Many research demonstrations and commercial applications have been developed from these efforts. Given an arbitrary image, the goal of face detection is to automatically locate a human face in an image or video, if it is present. Face detection in a general setting is a challenging problem due to the variability in scale, location, orientation (up-right, rotated), and pose (frontal, pro.le). Facial expression, occlusion, and lighting conditions also change the overall apprearance of faces. Yang et al. [Yang et al., 2002] summarize in their comprehensive survey the challenges associated with face detection:

- Pose. The images of a face vary due to the relative camera-face pose (frontal, 45 degree, pro.le, upside down), and some facial features (e.g., an eye or the nose) may become partially or wholly occluded.

- Presence or absence of structural components. Facial features such as beards, mustaches, and glasses may or may not be present and there is a great deal of variability among these components including shape, color, and size.

- Facial expression. The appearance of faces is directly affected by the facial expression of the persons.

-Occlusion. Faces may be partially occluded by other objects. In an image with a group of people, some faces may partially occlude other faces.

- Image orientation. Face images directly vary for different rotations about the camera’s optical axis.

-Imaging conditions. When the image is formed, factors such as lighting (spectra, source distribution and intensity) and camera characteristics (sensor response, lenses) affect the appearance of a face.