Document Analysis and Recognition - ICDAR 2023: 17th International Conference, San José, CA, USA, August 21–26, 2023, Proceedings, Part III (Lecture Notes in Computer Science)

This six-volume set of LNCS 14187, 14188, 14189, 14190, 14191 and 14192 constitutes the refereed proceedings of the 17 th International Conference on Document Analysis and Recognition, ICDAR 2021, held in San José, CA, USA, in August 2023. The 53 full papers were carefully reviewed and selected from 316 submissions, and are presented with 101 poster presentations. The papers are organized into the following topical sections: Graphics Recognition, Frontiers in Handwriting Recognition, Document Analysis and Recognition. Foreword Preface Organization Contents – Part III Posters: Document NLP Evaluation of Different Tagging Schemes for Named Entity Recognition in Handwritten Documents 1 Introduction 2 Framework 2.1 HTR and NER via a Coupled Model 2.2 Tagging Notation 2.3 Evaluation Metrics 3 Experimental Method 3.1 Datasets 3.2 Implementation Details 3.3 Obtained Results 4 Conclusions and Future Work References Analyzing the Impact of Tokenization on Multilingual Epidemic Surveillance in Low-Resource Languages 1 Introduction 2 Related Work 3 Epidemiological Event Dataset 4 Model and Task Selection 4.1 Language Models 4.2 Quality of the Pre-trained Tokenizer 5 Results and Analysis 5.1 Results and Error Analysis 5.2 Repairing Continued Entities 6 Conclusions and Future Work References DAMGCN: Entity Linking in Visually Rich Documents with Dependency-Aware Multimodal Graph Convolutional Network 1 Introduction 2 Related Works 2.1 GNN-Based Methods 2.2 BERT-Based Methods 3 Methodology 3.1 Graph Construction 3.2 Link Prediction 4 Experiments 5 Conclusion References Analyzing Textual Information from Financial Statements for Default Prediction 1 Introduction 2 Relevant Work 3 Financial Text Modelling 3.1 Financial Statement Processing Pipeline 3.2 Feature Extraction 4 Experimental Results 4.1 Datasets 4.2 Experiment Protocols 4.3 Analysis of Results 5 Conclusions References RealCQA: Scientific Chart Question Answering as a Test-Bed for First-Order Logic 1 Introduction 2 Background 2.1 Visual QA 2.2 Document Understanding 2.3 Chart-VQA 2.4 Logic Order and Reasoning 3 Dataset 3.1 RealCQA 3.2 Sampling Strategies for Dataset Evaluation 3.3 Evaluation Metrics 4 Experiments 4.1 Results 5 Conclusion References An Iterative Graph Learning Convolution Network for Key Information Extraction Based on the Document Inductive Bias 1 Introduction 2 Related Work 3 Methodology 3.1 Notation 3.2 Multi-modal Feature Extraction Module 3.3 Graph Module 3.4 Information Extraction Module 3.5 Loss 4 Experiments 4.1 Dataset 4.2 Implementation Details 4.3 Experimental Results 4.4 Ablation Studies 5 Conclusion References QuOTeS: Query-Oriented Technical Summarization 1 Introduction 2 Related Work 2.1 Query-Focused Summarization 2.2 High-Recall Information Retrieval 2.3 Interactive Query-Focused Summarization 3 Design Goals 4 System Design 4.1 Tutorial 4.2 Upload 4.3 Documents 4.4 Search 4.5 Explore 4.6 History 4.7 Results 5 Evaluation 5.1 Methodology 5.2 Participants 5.3 Research Instrument 5.4 Experimental Results 6 Discussion 6.1 Questionnaire Responses 6.2 Analysis of the Labels Collected During the User Study 6.3 Limitations 7 Conclusions and Future Work References A Benchmark of Nested Named Entity Recognition Approaches in Historical Structured Documents 1 Introduction 2 Related Works 2.1 Nested Named Entity Recognition Approaches 2.2 Named Entity Recognition on Noisy Inputs 2.3 Labels for NER on Highly Structured Documents 2.4 Conclusion 3 Considered Nested NER Approaches 3.1 State-of-the-Art Nested NER Approaches 3.2 A Hierarchical BERT-Based Transfer-Learning Approach Using Joint Labelling [M3] 4 Evaluation 4.1 Datasets 4.2 Experiments Summary 4.3 Tagging Formats 4.4 Pre-trained BERT-Based Models 4.5 Metrics 5 Results 5.1 Performance 5.2 Qualitative Analysis 6 Conclusion References ``Explain Thyself Bully'': Sentiment Aided Cyberbullying Detection with Explanation 1 Introduction 2 Related Works 2.1 Works on Monolingual Data 2.2 Works on Code-mixed Data 2.3 Works on Rationales: 2.4 Works on Sentiment Aware Multitasking 3 BullyExplain Dataset Development 3.1 Data Collection 3.2 Annotation Training 3.3 Main Annotation 3.4 Dataset Statistics 4 Explainable Cyberbullying Detection 4.1 Text Embedding Generation 4.2 Feature Extraction 4.3 mExCB :Multitask Framework for Explainable Cyberbullying Detection 4.4 Loss Function 5 Experimental Results and Analysis 5.1 Baselines Setup 5.2 Findings from Experiments 5.3 Comparison with SOTA 6 Error Analysis 7 Conclusion and Future Work References LayoutGCN: A Lightweight Architecture for Visually Rich Document Understanding 1 Introduction 2 Related Work 2.1 Pre-training Methods 2.2 Non-pre-training Methods 3 Approach 3.1 Document Modeling 3.2 Model Architecture 3.3 Downstream Tasks 4 Experiments 4.1 Datasets Description 4.2 Settings 4.3 Results 4.4 Case Study 4.5 Ablation Study 5 Conclusion References Topic Shift Detection in Chinese Dialogues: Corpus and Benchmark 1 Introduction 2 Related Work 2.1 Corpus 2.2 Topic Shift Detection in Dialogues 3 Corpus 3.1 Strengths 3.2 Annotation Guidelines 3.3 Data Source 3.4 Annotation Process 3.5 Annotation Results 4 Model 4.1 Knowledge Distillation 4.2 Hierarchical Contrastive Learning 4.3 Model Training 5 Experiments and Analysis 5.1 Experimental Settings 5.2 Experimental Results 5.3 Ablation Study 5.4 Analysis on Different Angles of Performance 5.5 Results on English TIAGE 5.6 Case Study and Error Analysis 6 Conclusion References Detecting Forged Receipts with Domain-Specific Ontology-Based Entities & Relations 1 Introduction 2 Related Work 3 Forged Receipt Dataset 4 Language Model Regression-based Approach 4.1 Model Description 4.2 Domain-specific Forged Receipt Input 5 Experiments 5.1 Hyperparameters 5.2 Results 6 Conclusions and Future Work References CED: Catalog Extraction from Documents 1 Introduction 2 Related Work 3 Dataset Construction 3.1 Processing & Annotation 3.2 Data Statistics 4 Transition-Based Catalog Extraction 4.1 Actions and Transition Process 4.2 Model Architecture 5 Experiments 5.1 Datasets 5.2 Evaluation Metrics 5.3 Baselines 5.4 Experiment Settings 5.5 Main Results 5.6 Analysis of Transfer Ability 5.7 Analysis on the Number of Training Data 5.8 Analysis on Different Depth 6 Conclusion and Future Discussion References A Character-Level Document Key Information Extraction Method with Contrastive Learning 1 Introduction 2 Related Work 2.1 Data-Sets 2.2 Pre-trained Language Model 2.3 Multi-modal Method 2.4 2D Grid-Based Method 3 Methodology 3.1 Character Level Encoder Module 3.2 Decoder Module with Label Feedback Mechanism 3.3 Similarity Module with Contrastive Learning 4 Experiment 4.1 Data-Sets 4.2 Evaluation Metrics 4.3 Results 4.4 Ablation Study 5 Conclusion References Multimodal Rumour Detection: Catching News that Never Transpired! 1 Introduction 2 Related Works 3 Dataset 3.1 Original Dataset 3.2 Dataset Extension 4 Problem Statement 5 Proposed Methodology 5.1 Embedding Generation 5.2 Attention-based Textual and Image Feature Extraction Module 5.3 Cross-modal Fusion Module (CMF) 5.4 Rumour Classification Module 6 Experiments, Results, and Analysis 6.1 Experimental Setting and Evaluation Metrics 6.2 Results and Discussion 6.3 Comparison with State of the Art and Other Techniques 6.4 Classification Examples and Error Analysis 7 Conclusions and Future Works References Semantic Triple-Assisted Learning for Question Answering Passage Re-ranking*-1pc 1 Introduction 2 Related Work 3 Methodology 3.1 Information Extraction-Based Coverage and Overlap Scores 3.2 Coverage Analysis 3.3 Overlap Analysis 3.4 Exploratory Analysis of Coverage and Overlap Scores 3.5 Semantic-Triple Assisted Learning Using Coverage/overlap Scores 4 Experiments 4.1 Datasets 4.2 Evaluation Metrics 4.3 Model Training 5 Results 5.1 Effect of Overlap Scores 5.2 Effect of Coverage Scores 5.3 Analyzing the Effect of 5.4 Comparison with the State-of-the-art 6 Discussion and Future Work 7 Conclusion References I-WAS: A Data Augmentation Method with GPT-2 for Simile Detection 1 Introduction 2 Related Work 2.1 Simile Detection 2.2 Text Data Augmentation 3 Task and Methodology 3.1 Formulation of Simile Detection 3.2 Simile Sentences Augmentation 4 Experiments Setup 4.1 Diverse Test Set 4.2 Dataset 4.3 Baselines 4.4 Settings 5 Results and Analysis 5.1 Results 5.2 Analysis 5.3 Augmented Sample 6 Discussion 7 Conclusion and Future Work References Information Redundancy and Biases in Public Document Information Extraction Benchmarks*-1pc 1 Introduction 2 Background 2.1 Related Work 2.2 Datasets 2.3 Problem Statement 3 Approach 3.1 Motivation 3.2 Resampling Datasets 3.3 Models 4 Experiments 4.1 Experimental Setup 4.2 Results on Original Datasets 4.3 Results on Resampled Datasets 5 Conclusion References Posters: Data and Synthesis On Web-based Visual Corpus Construction for Visual Document Understanding 1 Introduction 2 Background and Related Work 2.1 VDU Backbones 2.2 Visual Corpus Construction for VDU 3 Web-based Visual Corpus Builder 3.1 Annotation 3.2 Rendering with Various Fonts 4 Experiment and Analysis 4.1 Setup 4.2 Experimental Results 5 Discussion 5.1 Scale-up Using CommonCrawl 5.2 Webvicob with Augraphy 5.3 PCA Analysis 6 Conclusion References Ambigram Generation by a Diffusion Model 1 Introduction 2 Related Work 2.1 Diffusion Denoising Probabilistic Model (DDPM) 2.2 Letter Image Generation 3 Ambigram Generation by a Diffusion Model 3.1 Outline 3.2 Details 4 Evaluation of Ambigrams and Ambigramability 4.1 Evaluating Ambigrams by a Classifier 4.2 Ambigramability 5 Experimental Results 5.1 Experiment Setup 5.2 Qualitative Evaluations 5.3 Quantitative Evaluations 5.4 Comparative Studies 6 Conclusion References Analyzing Font Style Usage and Contextual Factors in Real Images 1 Introduction 2 Related Work 2.1 Font Style Embedding and Application 2.2 Multimodal Analysis of Fonts 2.3 Word Embedding and Its Multimodal Analysis 3 Analysis of Font Style Usage in Real Images 3.1 Experimental Strategy 3.2 Image Dataset 3.3 Text Detection and Recognition 3.4 Font Style Feature Extraction 3.5 Word Embedding 4 Experimental Results 4.1 Verification of the Trained Style Feature Extractor 4.2 Analysis of Font Style Usage in Scene Images 4.3 Analysis of Font Style Usage in Book Cover Images 5 Conclusion References CCpdf: Building a High Quality Corpus for Visually Rich Documents from Web Crawl Data 1 Introduction 2 Related Works 3 Collecting and Processing PDFs 3.1 Common Crawl 3.2 PDF Links Extraction 3.3 URL-Based Language Detection 3.4 Filtering Out Spam 3.5 PDF Data Download Methods 3.6 Born Digital Scanner 3.7 Born Digital Detection Heuristics 3.8 OCR Processing 3.9 Language Identification 3.10 Produced Index 4 Exploration of PDFs 5 Discussion 6 Conclusions References ESTER-Pt: An Evaluation Suite for TExt Recognition in Portuguese 1 Introduction 2 Related Work 3 Synthetic Datasets 3.1 ESTER-Pt Synthetic Text-based 3.2 ESTER-Pt Synthetic Image-based 4 ESTER-Pt Hybrid Image-Based PDFs 5 ESTER-Pt Real Image-Based PDFs 6 Experiments 6.1 Experimental Setup 6.2 Results 7 Conclusion References Augraphy: A Data Augmentation Library for Document Images 1 Introduction and Motivation 2 Related Work 2.1 Data Augmentation 2.2 Robustness Testing 3 Document Distortion, Theory and Technique 4 Augraphy 4.1 Augraphy Augmentations 4.2 The Library 5 Document Denoising with Augraphy 5.1 Model Architecture 5.2 Data Generation 5.3 Training Regime 5.4 Results 6 Robustness Testing 6.1 Character Recognition 6.2 Face Detection Robustness Testing 7 Conclusion and Future Work References TextREC: A Dataset for Referring Expression Comprehension with Reading Comprehension*-1pc 1 Introduction 2 Related Work 2.1 Referring Expression Comprehension Datasets. 2.2 Vision-Language Tasks with Text Reading Ability 3 TextREC Dataset 3.1 Images 3.2 Referring Expressions 3.3 Statistics and Analysis 4 Method 4.1 Language Attention Network 4.2 Text-Guided Matching Module 4.3 Learning Objective 5 Experiment 5.1 Dataset and Evaluation Protocol 5.2 Implementation Details 5.3 Performance of the Baselines on TextREC Dataset 5.4 Ablation Studies 5.5 Visualization Analysis 6 Conclusion References SIMARA: A Database for Key-Value Information Extraction from Full-Page Handwritten Documents 1 Introduction 2 Related Work 2.1 Datasets 2.2 Models for End-to-end Page Recognition 3 The SIMARA Dataset 3.1 Source of the Documents 3.2 Finding Aids 3.3 Description of the Handwritten Finding Aids 3.4 Finding Aids and Original Documents 3.5 Production of the Annotated Data 4 Experimental Results 4.1 Datasets 4.2 Key-Value Information Extraction Model 4.3 Metrics 4.4 Results 5 Conclusion References Diffusion Models for Document Image Generation 1 Introduction 2 Background 3 Methodology 3.1 Diffusion Model 3.2 Condition Encoder 3.3 Auto-Encoder 4 Experimental Setup 4.1 Evaluation Metric 4.2 Datasets 4.3 Implementation Details and Hyperparameters 5 Results and Analysis 5.1 Quantitative Results 5.2 Qualitative Results 6 Conclusion References Receipt Dataset for Document Forgery Detection 1 Introduction 2 Related Work 3 Dataset Building for Forged Receipts Detection 3.1 SROIE Dataset 3.2 Forgery Campaign 3.3 Post-processing 3.4 Dataset Description 4 Experiments - Baselines 4.1 Text Classification 4.2 Image Classification 4.3 Evaluation 4.4 Results 5 Conclusions and Perspectives References EnsExam: A Dataset for Handwritten Text Erasure on Examination Papers 1 Introduction 2 Related Work 2.1 Text Removal Benchmark 2.2 Text Removal Methods 3 EnsExam Dataset 3.1 Image Collection and Data Statistic 3.2 Annotation Details 4 Methodology 4.1 Model Architecture 4.2 Soft Stroke Mask and Stroke Normalization Loss 4.3 Training Objective 5 Experiments 5.1 Evaluation Metrics 5.2 Implementation Details 5.3 Quantitative and Qualitative Results 5.4 Extension of the Proposed Method on Scene Text Removal 6 Conclusion References MIDV-Holo: A Dataset for ID Document Hologram Detection in a Video Stream 1 Introduction 2 Dataset 3 Benchmark: Hologram Detection Method 3.1 Problem Statement 3.2 Proposed Algorithm 3.3 Experimental Section 3.4 Typical Error Cases 4 Conclusion References Author Index