Machine learning methods for reverse engineering of defective structured surfaces

Danksagung......Page 7
Abstract......Page 9
Contents......Page 10
List of Abbreviations......Page 13
1 Introduction......Page 14
 1.1 Machine Learning for Reverse Engineering......Page 17
 1.2 Organisation of this Thesis......Page 19
 1.3 Publications......Page 20
 2.1 Machine Learning Methods......Page 25
  2.1.1 Support Vector Machines......Page 26
  2.1.2 Neural Networks......Page 30
  2.1.3 Neural Texture Synthesis......Page 44
 2.2 Parametric Curves and Surfaces......Page 47
  2.2.1 The algorithm of de Casteljau......Page 48
  2.2.2 Bezier Curves......Page 49
  2.2.3 B-spline Curves......Page 50
  2.2.4 Tensor Product B-spline Surfaces......Page 56
  2.2.5 Tensor Product T-spline Surfaces......Page 58
3 Parametrization in Curve and Surface Approximation......Page 62
  3.1.1 Point Parametrization......Page 63
  3.1.2 Knot Placement......Page 64
 3.2 SVMs for Knot Placement in B-spline Curve Approximation......Page 65
  3.2.1 Geometric Features......Page 67
  3.2.2 Training Set Generation......Page 69
  3.2.3 Score-based Knot Placement......Page 71
  3.2.4 Results......Page 75
  3.3.1 Parametrization Approach......Page 79
  3.3.2 Neural Network Architecture......Page 82
  3.3.3 Training Process......Page 86
  3.3.4 Results......Page 87
  3.4.1 T-spline Surface Skinning......Page 92
  3.4.3 Results......Page 95
  3.5.1 Point Parametrization......Page 99
  3.5.2 Knot Selection......Page 101
 3.6 Conclusion......Page 104
4 Classification of Geometric Primitives in Point Clouds......Page 107
 4.1 Related works......Page 108
 4.2 Training Data Generation......Page 110
  4.2.1 Synthetic Primitive Patches......Page 111
  4.2.2 Volumetric Grid......Page 113
 4.3 Geometric Features for SVM-based Classification......Page 115
 4.4 Neural Network Architecture......Page 119
 4.5 Results and Discussion......Page 120
  4.5.1 Evaluation of Geometric Features......Page 121
  4.5.2 Comparison to Convolutional Neural Networks......Page 123
  4.5.3 Homogenization......Page 127
 4.6 Conclusion......Page 128
5 CNN Texture Synthesis for High-Resolution Image Inpainting......Page 131
 5.1 Related Works......Page 132
 5.2 Patch-based texture synthesis for image inpainting......Page 133
  5.2.1 Synthesis Loss......Page 134
  5.2.2 Two-Scale Resolution Synthesis......Page 135
  5.2.3 Reference Patch Look-up......Page 138
  5.2.4 Inpainting Approach......Page 140
 5.3 Results......Page 141
 5.4 Discussion......Page 148
 5.5 Conclusion......Page 151
6 Concluding Remarks & Outlook......Page 152
Bibliography......Page 156
 A.2 Runtime Remarks Section 5.3......Page 170