Document Analysis and Recognition - ICDAR 2023: 17th International Conference, San José, CA, USA, August 21–26, 2023, Proceedings, Part VI (Lecture Notes in Computer Science, 14192)

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 VI Posters: Scene Text Text Reading Order in Uncontrolled Conditions by Sparse Graph Segmentation 1 Introduction 2 Related Work 2.1 Reading Order Detection 2.2 Spatial, Image Features and Multi-Modality 3 Proposed Method 3.1 Strong Patterns of Reading Order 3.2 Model Architecture 3.3 Recovering Reading Order from Model Predictions 3.4 Data Labeling 3.5 Limitations 4 Experiments 4.1 Datasets and Evaluation Metrics 4.2 Model Setup 4.3 Baselines 4.4 Results 5 Conclusions and Future Work References TDAE: Text Detection with Affinity Areas and Evolution Strategies 1 Introduction 2 Related Works 3 Approach 3.1 Affinity Module 3.2 Label Generation 3.3 Network Design 3.4 Additional Fine-Tuning Step 3.5 Loss Function 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Ablation Study 4.4 Comparisons with Previous Methods 5 Conclusion and Future Work References Visual Information Extraction in the Wild: Practical Dataset and End-to-End Solution 1 Introduction 2 Related Works 2.1 VIE Datasets 2.2 VIE Methods 2.3 Contrastive Learning 3 POIE Dataset 3.1 Data Collection 3.2 Data Characteristics 3.3 Data Split and Evaluation Protocol 4 Our Method 4.1 Contrast-Guided Feature Adjustment Module 4.2 Information Extraction Module 4.3 Loss Function 5 Experiments 5.1 Implementation Details 5.2 Analysis of Our Proposed Dataset 5.3 The Comparisons of Our Method and SOTA 5.4 Ablation Studies 5.5 Limitation 6 Conclusion References Scene Text Recognition with Image-Text Matching-Guided Dictionary 1 Introduction 2 Related Work 2.1 Scene Text Recognition 2.2 Vision-Language Learning 3 Method 3.1 Overall Architecture 3.2 Resemblant Words Generation 3.3 Scene Image-Text Matching Module 3.4 Overall Objective Function 4 Experimental Results 4.1 Datasets 4.2 Training Setting 4.3 Comparison with the Ordinary Dictionary Method and the State-of-the-art 4.4 Ablation Study 5 Conclusion References E2TIMT: Efficient and Effective Modal Adapter for Text Image Machine Translation 1 Introduction 2 Preliminary 2.1 OCR Model 2.2 MT Model 3 Methodology 3.1 Embedding Modal Adapter 3.2 Sequential Modal Adapter 3.3 Training of Modal Adapter 3.4 Inference 4 Experiments 4.1 Datasets 4.2 Experimental Settings 4.3 Comparison of Various Text Image Translation Models 4.4 Performance of OCR and MT Models 4.5 Generalization of Modal Adapter on Various OCR and MT Combinations 4.6 Analysis on Model Size and Decoding Speed of TIMT Models 4.7 Comparison with Adapter Tuning 4.8 Analysis on the Amount of End-to-End TIMT Dataset 4.9 Hyper-parameter Analysis 5 Related Work 5.1 Text Image Translation. 5.2 Methods of Bridging Encoder and Decoder 6 Conclusion References Open-Set Text Recognition via Shape-Awareness Visual Reconstruction 1 Introduction 2 Related Works 2.1 Text Recognition 2.2 Open-Set Text Recognition 2.3 Visual Reconstruction 3 Our Method 3.1 Glyph Reconstruction 3.2 Cycle Classification 3.3 Optimization 4 Experiments 4.1 Implementation Details 4.2 Bench-Marking Protocols 4.3 Ablative Studies and Sensitivity Analysis 4.4 Openset Performance 4.5 Close-Set Performance 5 Limitation 6 Conclusion References Accelerating Transformer-Based Scene Text Detection and Recognition via Token Pruning 1 Introduction 2 Related Work 2.1 End-To-End Scene Text Recognition 2.2 Transformer Acceleration 3 Methodology 3.1 Architecture 3.2 Training Details 4 Experiments 4.1 Text Detection and Recognition Datasets 4.2 Token Pruning 5 Conclusions References Text Enhancement: Scene Text Recognition in Hazy Weather 1 Introduction 2 Related Work 2.1 Scene Text Recognition 2.2 Image Dehazing 3 Proposed Method 3.1 Overview of Text Enhancement Network 3.2 Parallelly Sequential Residual Block (PSRB) 3.3 Overall Loss Function 4 Experiments 4.1 Dataset 4.2 Implementation Details 4.3 Ablation Study 4.4 Comparison with State-of-the-Art Image Enhancement Methods 5 Conclusion References Reading Between the Lanes: Text VideoQA on the Road 1 Introduction 2 Related Work 2.1 VideoQA 2.2 VideoQA Involving Video Text 2.3 Scene Text VQA 3 RoadTextVQA Dataset 3.1 Data Collection 3.2 Data Statistics and Analysis 4 Baselines 4.1 Heuristic Baselines and Upper Bounds 4.2 M4C 4.3 Singularity 4.4 GenerativeImage2Text 5 Experiments and Results 5.1 Experimental Setup 5.2 Results 6 Conclusions References TPFNet: A Novel Text In-painting Transformer for Text Removal*-6pt 1 Introduction 2 Related Work 3 TPFNet: Our Proposed Network 3.1 Encoder Design 3.2 Decoder Design 3.3 Discriminator Design 3.4 Loss Functions Used in Training 3.5 Part-2: Image Generation 4 Experimental Platform 5 Experimental Results 5.1 Quantitative Results 5.2 Qualitative Results 6 Ablation Study 7 Conclusion References Author Index