Jiang / Hadid / Pang Deep Learning in Object Detection and Recognition

1. Auflage 2019
ISBN: 978-981-10-5152-4
Verlag: Springer Nature Singapore
Format: PDF
Kopierschutz: 1 - PDF Watermark

E-Book, Englisch, 237 Seiten

This book discusses recent advances in object detection and recognition using deep learning methods, which have achieved great success in the field of computer vision and image processing. It provides a systematic and methodical overview of the latest developments in deep learning theory and its applications to computer vision, illustrating them using key topics, including object detection, face analysis, 3D object recognition, and image retrieval. The book offers a rich blend of theory and practice. It is suitable for students, researchers and practitioners interested in deep learning, computer vision and beyond and can also be used as a reference book. The comprehensive comparison of various deep-learning applications helps readers with a basic understanding of machine learning and calculus grasp the theories and inspires applications in other computer vision tasks.

?Xiaoyue Jiang received her Ph.D. degree in Computer Science and Technology from Northwestern Polytechnical University in 2006. From Xiaoyue Jiang received her Ph.D. degree in Computer Science and Technology from Northwestern Polytechnical University in 2006. From 2006 to 2012, she has worked in Vrije University of Brussels (Belgium), University of Birmingham (UK) and Queen's University of Belfast (UK) as assistant and associated research fellow, respectively. She has worked as associated professor at Northwestern Polytechnical University since 2012. Her research interests includes computer vision, image processing and pattern recognition.  She has published more than 50 research papers and is currently senior fellow and secretary of Shaanxi Society of Image and Graphics. Abdenour Hadid is an adjunct professor at the Center for Machine Vision and Signal Analysis at University of Oulu. He is the chairman of the Pattern Recognition Society of Finland. His research interests include biometrics and facial image analysis, local descriptors, machine learning and human-machine interaction. He has authored over 140< articles in different forums and coauthored a very popular Springer Book on Computer Vision Using Local Binary Patterns in 2011. Yanwei Pang received his Ph.D. degree in Electronic Engineering from the University of Science and Technology of China (USTC) in 2004. Currently, he is a professor at the School of Electronic Information Engineering, Tianjin University, China. He is also the founding director of the Visual Pattern Analysis Laboratory of Tianjin University. His research interests include deep convolutional neural networks, pattern recognition, machine learning, computer vision and digital image processing. He has authored more than 100 scientific papers, 24 of which were published in IEEE Transactions. Eric Granger earned his Ph.D. in EE from the Poly-technique Montréal in 2001, and worked as a defense scientist at DRDC-Ottawa (1999-2001), and in R&D with Mitel Networks (2001-04). He joined the École de Technologie Supérieure (Université du Québec), Montreal, in 2004, where he is presently full professor and director of LIVIA, a research laboratory on computer vision and artificial intelligence. His research focuses on adaptive pattern recognition, machine learning, computer vision and computational intelligence. Xiaoyi Feng received her Ph.D. degree in Electronics and Information from Northwestern Polytechnical University in 2001. She is currently a professor and vice dean of the School of Electronics and Information, Northwestern Polytechnical University, and the vice director of the key laboratory of Ministry of Education 'Aerospace electronics information perception and photoelectric control'. Her research interests include image processing, pattern recognition, computer vision, radar imaging, embedded system design and applications.  She is the executive director of Shaanxi Society of Image and Graphics, and senior member of the China Society of Electronics. 

Autoren/Hrsg.

Jiang, Xiaoyue

Hadid, Abdenour

Pang, Yanwei

Granger, Eric

Feng, Xiaoyi

Inhaltsverzeichnis

1;Preface;5
2;Contents;9
3;Acronyms;13
4;An Overview of Deep Learning;17
4.1;1 Brief Introduction;17
4.2;2 Basic Types;19
4.2.1;2.1 Stacked Autoencoders (SAEs);20
4.2.2;2.2 Deep Belief Networks (DBNs);21
4.2.3;2.3 Convolutional Neural Networks (CNNs);22
4.2.4;2.4 Recurrent Neural Networks (RNNs);24
4.2.5;2.5 Generative Adversarial Nets (GANs);25
4.3;3 Practical Applications;27
4.3.1;3.1 Audio Data;27
4.3.2;3.2 Image Data;28
4.3.3;3.3 Text Data;29
4.4;4 Existing Challenges;30
4.4.1;4.1 Theory Challenges;30
4.4.2;4.2 Engineering Challenges;31
4.5;5 Conclusions;31
4.6;References;32
5;Deep Learning in Object Detection;35
5.1;1 Introduction;35
5.2;2 The CNN Architectures of Object Detection;37
5.2.1;2.1 Two-Stage Methods for Deep Object Detection;37
5.2.2;2.2 One-Stage Methods for Deep Object Detection;46
5.3;3 Pedestrian Detection;50
5.3.1;3.1 Handcrafted Feature-Based Methods for Pedestrian Detection;50
5.3.2;3.2 CNN-Based Methods for Pedestrian Detection;56
5.4;4 Challenges of Object Detection;62
5.4.1;4.1 Scale Variation Problem;62
5.4.2;4.2 Occlusion Problem;68
5.4.3;4.3 Deformation Problem;69
5.5;References;70
6;Deep Learning in Face Recognition Across Variations in Pose and Illumination;74
6.1;1 Introduction;74
6.2;2 Pose-Invariant Face Recognition;76
6.2.1;2.1 Invariant Representation;76
6.2.1.1;2.1.1 Engineering Designed Features;76
6.2.1.2;2.1.2 Learning-Based Features;79
6.2.2;2.2 Synthesis-Based Methods;87
6.2.2.1;2.2.1 2D-Based Synthesis Methods;87
6.2.2.2;2.2.2 3D-Based Synthesis Methods;89
6.3;3 Illumination-Invariant Face Recognition;89
6.3.1;3.1 Image Processing-Based Methods;89
6.3.2;3.2 Invariant Feature-Based Methods;91
6.3.3;3.3 Illumination Model-Based Method;93
6.4;4 Multi-stream Convolutional Neural Networks;96
6.4.1;4.1 Root Convolutional Layer;96
6.4.2;4.2 Multi-hierarchical Local Feature;98
6.4.3;4.3 Training;99
6.5;5 Experiments;99
6.5.1;5.1 Dataset;99
6.5.2;5.2 Recognition Across Poses and Illumination;100
6.6;6 Conclusion;101
6.7;References;102
7;Face Anti-spoofing via Deep Local Binary Pattern;106
7.1;1 Introduction;106
7.2;2 Related Work;108
7.3;3 Proposed Method;110
7.3.1;3.1 CNN Training;111
7.3.2;3.2 Color Spaces;111
7.3.3;3.3 Local Binary Pattern (LBP);111
7.3.4;3.4 Concatenating the LBP;113
7.3.5;3.5 Classification;114
7.4;4 Experimental Data and Setup;114
7.4.1;4.1 Experimental Data;114
7.4.1.1;4.1.1 Replay-Attack;114
7.4.1.2;4.1.2 CASIA-FA;115
7.4.2;4.2 Experimental Setups;116
7.4.2.1;4.2.1 Evaluation Protocol;116
7.4.2.2;4.2.2 Data Processing;116
7.4.2.3;4.2.3 Intra Test and Cross Test;117
7.5;5 Results and Discussion;117
7.5.1;5.1 Impact of Different Color Spaces;117
7.5.1.1;5.1.1 Intra Test;117
7.5.1.2;5.1.2 Cross Test;120
7.5.2;5.2 Concatenating Different Color Spaces;120
7.5.2.1;5.2.1 Intra Test;120
7.5.2.2;5.2.2 Cross Test;121
7.5.3;5.3 Comparison Against State-of-the-Art Methods;121
7.6;6 Conclusion;123
7.7;References;125
8;Kinship Verification Based on Deep Learning;127
8.1;1 Introduction;127
8.2;2 Related Work;128
8.2.1;2.1 Methods Based on Features;129
8.2.2;2.2 Methods Using Metric Learning;130
8.2.3;2.3 Other Methods;130
8.3;3 Image-Based Kinship Verification;131
8.3.1;3.1 Methodology;131
8.3.2;3.2 Experimental Analysis;135
8.4;4 Video-Based Kinship Verification;136
8.4.1;4.1 Methodology;137
8.4.2;4.2 Experimental Analysis;140
8.5;5 Conclusion;144
8.6;References;144
9;Deep Learning Architectures for Face Recognition in VideoSurveillance;147
9.1;1 Introduction;148
9.2;2 Background of Video-Based FR Through Deep Learning;149
9.3;3 Deep Learning Architectures for FR in VS;151
9.3.1;3.1 Deep CNNs Using Triplet-Loss;151
9.3.1.1;3.1.1 Cross-Correlation Matching CNN;151
9.3.1.2;3.1.2 Trunk-Branch Ensemble CNN;155
9.3.1.3;3.1.3 HaarNet;156
9.3.2;3.2 Deep CNNs Using Autoencoder;161
9.4;4 Performance Evaluation;164
9.5;5 Conclusion and Future Directions;165
9.6;References;166
10;Deep Learning for 3D Data Processing;169
10.1;1 Introduction;170
10.2;2 Related Works;172
10.2.1;2.1 Knowledge-Based 3D Local Features;172
10.2.2;2.2 Deep Learning for 3D Shapes With Raw Features;172
10.2.3;2.3 Restricted Boltzmann Machines (RBM) and Deep Belief Network (DBN);173
10.2.4;2.4 Convolutional RBM (CRBM) and Convolutional DBN (CDBN);174
10.2.5;2.5 Details of CRBM and CDBN;174
10.3;3 Circle Convolutional RBM (CCRBM);176
10.3.1;3.1 Circle Convolution;176
10.3.2;3.2 Projection Distance Distribution (PDD);180
10.3.3;3.3 Example of Circle Convolution and PDD Computation;180
10.3.4;3.4 Elimination of the Initial Location Ambiguity;181
10.3.5;3.5 The Structure of CCRBM;183
10.3.6;3.6 Circle Convolutional DBN (CCDBN);186
10.4;4 Experimental Setup;187
10.4.1;4.1 Global Shape Retrieval;187
10.4.2;4.2 Partial Shape Retrieval;188
10.4.3;4.3 Shape Correspondence;188
10.4.4;4.4 The Setup of Parameters for CCRBM;189
10.5;5 Results and Analysis;193
10.5.1;5.1 Global Shape Retrieval;193
10.5.2;5.2 Partial Shape Retrieval;194
10.5.3;5.3 Shape Correspondence;195
10.5.4;5.4 Significance and Analysis;196
10.5.5;5.5 Limitation and Future Work;197
10.6;6 Conclusion;197
10.7;References;198
11;Deep Learning-Based Descriptors for Object Instance Search;202
11.1;1 Introduction;203
11.2;2 Related Work;204
11.3;3 Compact Invariant Deep Descriptors;207
11.3.1;3.1 Restricted Boltzmann Machine for Hashing;208
11.3.1.1;3.1.1 Method;208
11.3.1.2;3.1.2 Evaluation Framework;212
11.3.1.3;3.1.3 Experimental Results;213
11.3.2;3.2 Dual-Margin Siamese Fine-Tuning;215
11.3.2.1;3.2.1 Method;217
11.3.2.2;3.2.2 Evaluation Framework;219
11.3.2.3;3.2.3 Experimental Results;219
11.3.3;3.3 Nested Invariance Pooling;221
11.3.3.1;3.3.1 I-Theory in a Nutshell;221
11.3.3.2;3.3.2 CNNs Are I-Theory Compliant Networks;222
11.3.3.3;3.3.3 Multigroup-Invariant CNN Descriptors;222
11.3.3.4;3.3.4 Evaluation Framework;225
11.3.3.5;3.3.5 Experimental Results;226
11.3.4;3.4 Hashing with Invariant Descriptors;229
11.4;4 Conclusions and Future Works;232
11.5;References;233