Graph-based keyword spotting

Description

Contents Preface List of Figures List of Tables 1. Introduction 1.1 Keyword Spotting (KWS) as a Scientific Discipline 1.2 Limitations 1.3 Research Questions 1.4 Contributions 1.5 Outline and Organisation of the Book 2. Related Work 2.1 A Taxonomy of KWS Systems 2.2 Image Preprocessing 2.3 Formal Representation 2.3.1 Statistical Representation 2.3.1.1 Handwriting Descriptors 2.3.1.2 Texture and Shape Descriptors 2.3.1.3 Encoding and Embedding Approaches 2.3.2 Structural Representation 2.4 Querying 2.4.1 Template-based Approaches for KWS 2.4.1.1 Statistical Representation 2.4.1.2 Structural Representation 2.4.2 Learning-based Approaches for KWS 2.4.2.1 Statistical Representation 2.4.2.2 Structural Representation 2.5 Summary 3. Datasets 3.1 Historical Manuscripts 3.1.1 George Washington (GW) 3.1.2 Parzival (PAR) 3.1.3 Alvermann Konzilsprotokolle (AK) 3.1.4 Botany (BOT) 3.2 Preprocessing the Datasets 3.2.1 Gaussian Filtering 3.2.2 Binarisation 3.2.3 Segmentation 3.2.4 Skew Correction 3.2.5 Morphological Filtering 3.2.6 Skeletonisation 3.3 Summary 4. Graph Representations 4.1 Introduction and Basic Definitions 4.2 Graph-Based Pattern Recognition 4.3 From Handwriting to Graphs 4.3.1 Keypoint-Based Graphs 4.3.2 Grid-Based Graphs 4.3.3 Projection-Based Graphs 4.3.4 Split-Based Graphs 4.4 Preliminary Experimental Evaluation 4.5 Summary 5. Graph Matching 5.1 Overview and Broader Perspective 5.2 Graph Matching for KWS 5.2.1 Exact Graph Matching 5.2.2 Inexact Graph Matching 5.2.3 Formal Requirements 5.3 Graph Edit Distance 5.3.1 Suboptimal Algorithms 5.3.2 Bipartite Graph Edit Distance (BP) 5.3.3 Hausdor Edit Distance (HED) 5.3.3.1 Context-aware Hausdor Edit Distance (CED) 5.3.4 Bipartite Graph Edit Distance 2 (BP2) 5.4 Polar Graph Dissimilarity (PGD) 5.4.1 Polar Segmentation of Graphs 5.4.2 Histogram-Based Representation 5.4.3 Histogram-Based Dissimilarity Measure 5.5 Summary 6. Graph-Based Keyword Spotting 6.1 Overview of our Framework for KWS 6.2 From Graph Edit Distance (GED) to Retrieval Indices 6.2.1 Cost Model for Keyword Spotting 6.2.2 Retrieval Indices 6.3 Extensions of the Basic Framework 6.3.1 Quadtree Segementations 6.3.2 Fast Rejection Methods 6.3.3 Ensemble Methods 6.4 Evaluation Metric 6.5 Summary 7. Experiments 7.1 Overview of Experimental Evaluation 7.2 Experimental Setup 7.2.1 Validation of Parameters 7.3 The Baseline KWS System Using BP 7.4 Speeding Up the Baseline KWS System 7.4.1 KWS System Based on BP with Quadtree Segmentations 7.4.2 KWS System Based on BP with Fast Rejection Methods 7.4.3 KWS System Based on PGD 7.5 KWS System Based on HED 7.5.1 Hausdor Edit Distance 7.5.2 Context-aware Hausdor Edit Distance 7.6 KWS System Based on BP2 7.7 Graph-Based Ensembles for KWS 7.7.1 Ensemble Methods Based on BP 7.7.2 Ensemble Methods Based on HED, CED, and BP2 7.8 Cross-Evaluation of Graph-Based Systems 7.9 Quantitative and Qualitative Summary of Graph-Based Systems 7.9.1 Quantitative Summary 7.9.2 Qualitative Summary 7.10 Comparison of Our Framework with Reference Systems 7.10.1 Reference Systems 7.10.1.1 Template-based Reference Systems 7.10.1.2 Learning-based Reference Systems 7.10.2 Comparison with Template-Based Reference Systems 7.10.3 Comparison with Learning-Based Reference Systems 7.11 Summary 8. Conclusion & Future Work 8.1 Conclusion 8.2 Future Work 9. Visualisation of Graph Representations 9.1 Visualisation of GW 9.2 Visualisation of PAR 9.3 Visualisation of AK 9.4 Visualisation of BOT 10. Optimisation of the Parameters 11. Ensemble Methods 11.1 Ensemble Methods Based on HED 11.2 Ensemble Methods Based on CED 11.3 Ensemble Methods Based on BP2 Bibliography Index