PhD Thesis Semi-Supervised Ensemble Methods for Computer Vision

More documents

Recommendations

Info

2 Chapter 1. Introduction Figure 1.1: The eyes are the most important senses for a frog. 1982] and states the scientific field where the thesis at hand is placed. Why is it worth doing research on computer vision? The answer to this question can be split up into two parts: First, from a researcher’s perspective and due to the human pioneering spirit, i.e., simply because we want to find out if we can do it. Second, because there exist a huge amount of potential applications. Some of these are optical character recognition (OCR), web-based image search, industrial machine inspection, biomedical imaging, surveillance, 3D model building and photogrammetry, automotive safety, biometrics, robotics, etc.. See also David Lowe’s website of industrial vision applications 1 . Overall, computer vision already now comprises a multi-billion dollar market with expected steady growth. Vision is hard The human visual system rapidly and effortlessly recognizes a large number of diverse objects despite large variations in the object’s position, pose, lighting and background clutter. Additionally, we can easily segment an object, analyze its shape and track it. Building computational systems that are able to achieve the same performance is extremely hard. One of the reasons for this difficulty is that “reverse engineering the brain” in order to emulate it on machines is very complicated. Although cognitive science has made large progress in understanding the brain’s solution to perform visual tasks (see above), we are still far from fully understanding how human perception works. Additionally, unlike humans, we provide computers with digital data and from a machine’s perspective, an image is nothing else than a matrix of numbers. The size and the quality of the matrices may vary a lot. Hence, the way how computer vision is done today can be also interpreted as searching for useful information in matrices and researchers still argue if this is the right path to follow. Computer vision is an inverse discipline, that is we have to find a solution for a problem where we get provided by an insufficient amount of information; and inverse problems are typically ill-posed, i.e., there does not a exist a unique solution [Hadamard, 1 http:://www.cs.ubc.ca/spider/lowe/vision.html (01.04.2010)
3 1902]. Hence, computer vision has often to make use of probabilistic approaches and the success of the methods relies on how good our probabilistic models are and on the quality of information with which we feed them. Finally, the human visual system is able to cope with large scales of data and thus acts as a massively parallel computer, comprised of billions of elements. Hence, when doing vision on machines we also have to cope with a huge amount of data, which requires both large computational power and highly efficient algorithms in order to deliver results in acceptable time if not real-time. Altogether, these are big challenges both for algorithm designers and hardware developers. The role of Machine Learning and Training Samples Computer vision can be subdivided into several disciplines, such as, structure from motion, segmentation, reconstruction, action recognition, etc. [Szeliski, 2010]. In this thesis, we are mainly dealing with visual recognition and classification as well as object tracking. A typical machine perception or pattern recognition system can be subdivided into the following steps (See also Duda [Duda et al., 2001]): 1. Sensing 2. Segmentation 3. Feature extraction and selection 4. Classification 5. Post-processing The success of a recognition system clearly depends on the quality of either of these steps; however, especially good representations in terms of features and training accurate classifiers mostly determine the overall quality of a recognition system. Hence, in recent years, further developments in terms of representation and novel machine learning algorithms have brought the highest accuracy improvements. In fact, especially machine learning techniques become increasingly relevant for computer vision and, according to this observation, this is also the reason why in this thesis we mainly focus on the development and application of learning and classification algorithms for computer vision. The task of a machine learning algorithm is to, based on provided training samples, train classifiers that predict the labels of samples that have not been observed during training. In practice, both the training samples and their corresponding labels are provided by a human labeler. The learners are thus called supervised methods. A lot of research has been focused on developing new classifiers and learning algorithms. If enough and proper training samples exist, these approaches can obtain very high recognition and classification performances. However, one fact that has been often ignored or overlooked in
Page 1: PhD Thesis Semi-Supervised Ensemble
Page 5: Statutory Declaration I declare tha
Page 8 and 9: Most of all, I would like to thank
Page 10 and 11: learning. Finally, we hypothesize t
Page 12 and 13: sten Teil dieser Arbeit schlagen wi
Page 14 and 15: ii CONTENTS 3.6 Graph-based Methods
Page 16 and 17: iv CONTENTS 10 Conclusion 137 10.1
Page 18 and 19: vi LIST OF FIGURES 4.8 Performance
Page 20 and 21: viii LIST OF FIGURES 9.7 Comparison
Page 22 and 23: x LIST OF FIGURES
Page 24 and 25: xii LIST OF TABLES 8.2 Results and
Page 26 and 27: xiv LIST OF ALGORITHMS
Page 30 and 31: 4 Chapter 1. Introduction the liter
Page 32 and 33: 6 Chapter 1. Introduction 1.1 Contr
Page 34 and 35: 8 Chapter 1. Introduction
Page 36 and 37: 10 Chapter 2. Preliminaries and Not
Page 52 and 53: 26 Chapter 3. Overview of Semi-Supe
Page 70 and 71: 44 Chapter 4. SemiBoost and Visual
Page 78 and 79:
52 Chapter 4. SemiBoost and Visual
Page 80 and 81:
Page 82 and 83:
Page 84 and 85:
Page 86 and 87:
Page 88 and 89:
62 Chapter 5. On-line Semi-Supervis
Page 90 and 91:
Page 92 and 93:
Page 94 and 95:
Page 96 and 97:
Page 98 and 99:
Page 100 and 101:
Page 102 and 103:
Page 104 and 105:
Page 106 and 107:
Page 108 and 109:
Page 110 and 111:
84 Chapter 6. Semi-Supervised Rando
Page 112 and 113:
Page 114 and 115:
Page 116 and 117:
Page 118 and 119:
Page 120 and 121:
Page 122 and 123:
Page 124 and 125:
Page 126 and 127:
100 Chapter 7. On-line Semi-Supervi
Page 128 and 129:
102 Chapter 7. On-line Semi-Supervi
Page 130 and 131:
104 Chapter 8. Multiple Instance Le
Page 132 and 133:
Page 134 and 135:
Page 136 and 137:
Page 138 and 139:
Page 140 and 141:
Page 142 and 143:
116 Chapter 9. Visual Object Tracki
Page 144 and 145:
Page 146 and 147:
Page 148 and 149:
Page 150 and 151:
Page 152 and 153:
Page 154 and 155:
Page 156 and 157:
Page 158 and 159:
Page 160 and 161:
Page 162 and 163:
Page 164 and 165:
138 Chapter 10. Conclusion As many
Page 166 and 167:
140 Chapter 10. Conclusion positive
Page 168 and 169:
142 Chapter 10. Conclusion
Page 170 and 171:
144 Chapter A. Publications (8) Mar
Page 172 and 173:
146 Chapter A. Publications
Page 174 and 175:
148 Chapter B. Acronyms SVM Support
Page 176 and 177:
150 BIBLIOGRAPHY [Balcan et al., 20
Page 178 and 179:
152 BIBLIOGRAPHY [Chapelle and Zien
Page 180 and 181:
154 BIBLIOGRAPHY [Gall and Lempinsk
Page 182 and 183:
156 BIBLIOGRAPHY [Leistner et al.,
Page 184 and 185:
158 BIBLIOGRAPHY [Nigam et al., 200
Page 186 and 187:
160 BIBLIOGRAPHY [Shalev-Shwartz, 2
Page 188 and 189:
162 BIBLIOGRAPHY [Xu et al., 2009]
show all

PhD Thesis Semi-Supervised Ensemble Methods for Computer Vision

Create successful ePaper yourself

Delete template?

Save as template?