Pattern Recognition. ICPR International Workshops and Challenges: Virtual Event, January 10-15, 2021, Proceedings, Part VII (Image Processing, Computer Vision, Pattern Recognition, and Graphics)

Foreword by General Chairs
Preface
Challenges
ICPR Organization
Contents – Part VII
PATCAST - International Workshop on Pattern Forecasting
International Workshop on Pattern Forecasting
Organization
General Chairs
Adaptive Future Frame Prediction with Ensemble Network
1 Introduction
2 Related Works
3 Proposed Ensemble Network
3.1 Overall Network Architecture
3.2 Future Frame Prediction Network
3.3 Network Training
4 Experiments
4.1 Experimental Environment
4.2 Performance of Future Frame Prediction Network Architecture
4.3 Future Frame Prediction on Online-Updating
5 Conclusions
References
Rain-Code Fusion: Code-to-Code ConvLSTM Forecasting Spatiotemporal Precipitation
1 Introduction
1.1 Rain-Code Fusion Based Precipitation Forecasting
1.2 Related Works and Papers
1.3 Stretch Predictability Using Code-to-Code Forecasting
2 Precipitation Forecasting Method
2.1 Multi-frame Rain-Code Against Single-Frame Sequence
2.2 Multi-frame-Based Forecasting Precipitation Method
2.3 Hourly Accuracy Indices for Rain-Code Forecasting
3 Applied Results
3.1 Training and Test Datasets for Matrix Prediction
3.2 Feasibility Study of Rain-Code Forecasting and Computing Accuracy
4 Rain-Code Size Sensitivity Studies
4.1 Rain-Code Training Results and Accuracy Indices for Different Number of Frames
4.2 Rain-Code Based Forecasting Results and Accuracy Using 2 × 2 Frames
4.3 Rain-Code Based Forecasting Results and Accuracy Using 3 × 4 Frames with Padding 1 × 4 mask
5 Concluding Remarks
5.1 Multi-frame Based Code-to-Code Spatiotemporal Forecasting
5.2 Future Extend Forecasting Range for Dam Inflow Prediction
References
PATRECH2020 - II International Workshop on Pattern Recognition for Cultural Heritage
PatReCH 2020 - 2nd International Workshop on Pattern Recognition for Cultural Heritage
Workshop Description
Organization
Workshop Chairs
Program Committee
Using Graph Neural Networks to Reconstruct Ancient Documents
1 Introduction
2 Related Works
3 Automatic Reconstruction Using a ConvGNN
3.1 Creation of a Ground Truth Dataset
3.2 Model Architecture
3.3 Experiments and Results
4 Interactive Visualization of Assembly Proposals
4.1 Graphical Interface
4.2 Examples of Image Reconstructions
4.3 Reconstructions from Multiple Images
5 Discussion and Conclusion
References
AnCoins: Image-Based Automated Identification of Ancient Coins Through Transfer Learning Approaches
1 Introduction
2 Background and Related Works
3 TheAnCoins-12 Dataset
4 Experiments
5 AnCoins Web-Based System
6 Discussion: Features Identification in Coins
7 Conclusions and Future Directions
References
Subjective Assessments of Legibility in Ancient Manuscript Images - The SALAMI Dataset
1 Introduction
2 Study Design
2.1 Test Images
2.2 Test Method
2.3 Rating Scale
2.4 Test Environment
2.5 Order of Presentation
2.6 Participants
3 Experiment Conduction
4 Evaluation
4.1 Participant Characteristics and Agreement
4.2 Systematic Effects and Sources of Variation
4.3 Spatial Distribution of Variability
5 Dataset Description and Validity
6 Conclusion
References
Can OpenPose Be Used as a 3D Registration Method for 3D Scans of Cultural Heritage Artifacts
1 Introduction
2 Related Work
2.1 Traditional Approaches
2.2 Deep Learning Approaches
3 Method Description
3.1 3D Scanning System Components
3.2 Providing Data for a Coarse 3D Registration - OpenPose
3.3 Providing Data for a Fine 3D Registration - ICP
4 Results and Discussion
4.1 Limitations
5 Conclusion
References
Survey on Deep Learning-Based Kuzushiji Recognition
1 Introduction
2 Representative Research on Kuzushiji Recognition
3 Datasets
3.1 Kuzushiji Dataset
3.2 Electronic Kuzushiji Dictionary Database
3.3 Wooden Tablet Database
4 Benchmarks
4.1 PRMU Algorithm Contest
4.2 Kaggle Competition
5 Activities Related to Kuzushiji Recognition
6 Future Studies on Kuzushiji Recognition
7 Conclusion
References
Stylistic Classification of Historical Violins: A Deep Learning Approach
1 Introduction
2 Related Works
3 Dataset
4 Network Architecture
5 Experimental Results
6 Conclusions
References
Text Line Extraction Using Fully Convolutional Network and Energy Minimization
1 Introduction
2 Related Work
3 Datasets
3.1 VML-AHTE
3.2 Diva-HisDB
3.3 VML-MOC
4 Method
4.1 Text Line Detection Using FCN
4.2 Text Line Extraction Using EM
5 Experiments
5.1 ICDAR 2013 Line Segmentation Evaluation Metrics
5.2 ICDAR 2017 Line Segmentation Evaluation Metrics
5.3 Results on VML-AHTE Dataset
5.4 Results on VML-MOC Dataset
5.5 Results on DIVA-HisDB Dataset
5.6 Discussion
6 Conclusion
References
Handwriting Classification of Byzantine Codices via Geometric Transformations Induced by Curvature Deformations
1 Introduction
1.1 Relation of the Proposed Methodology with State of the Art
2 The Introduced Methodology for Matching and Comparing Letters’ Shapes
2.1 Fundamental Entities and Hypotheses
2.2 Geometrical Setting of the Letters Similarity Problem
2.3 Derivation of the Implicit Curvature Deformation Rule
2.4 Optimal Affine Registration of the Letters’ Shapes
2.5 Evaluation of the Joint Implicit Family Hypothesis
3 Statistical Classification of Documents into Writers
4 Identification of the Writer of Important Historical Documents
5 Conclusion
References
Visual Programming-Based Interactive Analysis of Ancient Documents: The Case of Magical Signs in Jewish Manuscripts
1 Introduction
2 Background and Related Work
2.1 Introduction to charaktêres
2.2 Visual Properties
2.3 Computational Analysis
3 System Design
3.1 User Requirements
3.2 Approach
4 Analyzing charaktêres with AMAP
4.1 Interactive Exploration of Toolchains
4.2 Limitations
5 User Evaluation
5.1 Participants
5.2 Design
5.3 Methods and Results
6 Conclusion and Future Work
References
Quaternion Generative Adversarial Networks for Inscription Detection in Byzantine Monuments
1 Introduction
2 Related Work
3 Elements of Quaternions
4 Quaternionic Convolutional Neural Networks
5 Proposed Model
6 Experimental Results
6.1 Dataset
6.2 Experiments
7 Conclusion and Future Work
References
Transfer Learning Methods for Extracting, Classifying and Searching Large Collections of Historical Images and Their Captions
1 Introduction
2 Related Work
2.1 Applications and Data Analysis Methods with Historical Data
3 Information Retrieval
3.1 Image Extractions
3.2 Text Extraction
4 Data Analysis
4.1 Image Similarity
4.2 Caption Similarity
5 Research Tool
5.1 Word Lookup
5.2 Similarity Calculation
6 Conclusions and Future Work
References
Deep Learning Spatial-Spectral Processing of Hyperspectral Images for Pigment Mapping of Cultural Heritage Artifacts
1 Introduction
2 Dataset and Methodology
2.1 Labelled Reference Data of the Gough Map
2.2 Input and Output (I/O) of the Neural Network
2.3 3D-SE-ResNet Architecture
3 Experimental Results
3.1 Experimental Results from the Dataset Aspect
3.2 Experimental Results from Framework Aspect
4 Conclusions
References
Abstracting Stone Walls for Visualization and Analysis
1 Introduction
2 The Proposed Technique
3 Two Case Studies
4 Conclusions and Future Work
References
PapyRow: A Dataset of Row Images from Ancient Greek Papyri for Writers Identification
1 Introduction
2 Description of the Original Dataset
3 Image Enhancement
3.1 Background Smoothing
3.2 Line Resizing
3.3 Image Rotation
3.4 Row Labeling
4 Description of the Final Dataset and Conclusion
References
Stone-by-Stone Segmentation for Monitoring Large Historical Monuments Using Deep Neural Networks
1 Introduction
2 Related Work
3 Methods
3.1 Orthomosaic Map
3.2 Edge Detection and Thresholding Methods
3.3 Deep Learning Methods
4 Experimental Results
4.1 Dataset
4.2 Comparative Tests
5 Conclusion and Future Work
References
A Convolutional Recurrent Neural Network for the Handwritten Text Recognition of Historical Greek Manuscripts
1 Introduction
2 Related Work
3 Proposed Methodology
3.1 Architecture
3.2 Training
4 Experimental Results
4.1 Datasets
4.2 Ablation Study
4.3 Error Analysis in the `EPARCHOS' Dataset
5 Conclusions
References
MCCNet: Multi-Color Cascade Network with Weight Transfer for Single Image Depth Prediction on Outdoor Relief Images
1 Introduction
2 Related Work
3 Proposed Method
3.1 Dataset
3.2 Multi-Color Cascade Network with Weight Transfer
4 Experiments and Results
4.1 Optimal Network Architecture and Comparison on Prambanan RRD Dataset
4.2 Comparison on NYU Depth V2 Dataset
5 Conclusion
References
Simultaneous Detection of Regular Patterns in Ancient Manuscripts Using GAN-Based Deep Unsupervised Segmentation
1 Introduction
2 Related Work
3 Overview of Proposed Method
3.1 Work Methodology
3.2 Learning Networks
4 Experiments
4.1 Page Segmentation
4.2 Ornament Segmentation
4.3 Character Segmentation
5 Conclusion
References
A Two-Stage Unsupervised Deep Learning Framework for Degradation Removal in Ancient Documents
1 Introduction
2 Related Work
3 Work Methodology
3.1 Stage I - Data Augmentation Framework
3.2 Stage II Convolutional Neural Network-Based Document Binarization
3.3 Datasets
4 Result and Analysis
5 Conclusion
References
Recommender System for Digital Storytelling: A Novel Approach to Enhance Cultural Heritage
1 Introduction
2 Background
2.1 Recommender Systems
2.2 Context-Awareness
2.3 Digital Storytelling
3 The Proposed Approach
3.1 Recommendation Module
4 Experimental Results
5 Conclusion and Future Works
References
A Contextual Approach for Coastal Tourism and Cultural Heritage Enhancing
1 Introduction
2 System Architecture
2.1 Context Representation
2.2 Data Management and Representation
2.3 Inferential Engines
3 Experimental Results
4 Conclusion and Future Works
References
A Comparison of Character-Based Neural Machine Translations Techniques Applied to Spelling Normalization
1 Introduction
2 Related Work
3 Normalization Approaches
3.1 Character-Based SMT
3.2 Character-Based NMT
4 Experiments
4.1 Systems
4.2 Corpora
4.3 Metrics
5 Results
5.1 In-depth Comparison
6 Conclusions and Future Work
References
Weakly Supervised Bounding Box Extraction for Unlabeled Data in Table Detection
1 Introduction
2 Related Work
3 Proposed Method and Dataset
3.1 Preprocessing
4 Experiment and Results
5 Conclusion
References
Underground Archaeology: Photogrammetry and Terrestrial Laser Scanning of the Hypogeum of Crispia Salvia (Marsala, Italy)
1 Introduction
1.1 Machine Learning and Advances in Archaeological Practice
1.2 Digital Methods for Hypogeal Contexts
1.3 Digital Methods for Hypogeal Contexts in Sicily
2 Materials
2.1 The Hypogeum
2.2 The Inscription
2.3 The Frescoes
3 Methods
3.1 Results
3.2 Discussion
4 Conclusion
References
PRAConBE - Pattern Recognition and Automation in Construction and the Built Environment
Preface
Organization
Workshop Chairs
Invited Speaker
Program Committee
Automatic MEP Component Detection with Deep Learning
1 Introduction
2 Related Work
2.1 Mathematical Algorithms
2.2 Machine Learning
2.3 Deep Learning
3 Methodology
3.1 Dataset
3.2 Model Training and Validation
3.3 Dataset Augmentation
4 Experiments and Results
4.1 Dataset
4.2 Measuring Performance
4.3 360 Image MEP Detector
4.4 Standard Image MEP Detector
4.5 Cross-Format Testing
5 Conclusion
5.1 Summary and Limitations
5.2 Future Research
References
Mixed Reality-Based Dataset Generation for Learning-Based Scan-to-BIM
1 Introduction
2 Automation in Scan-to-BIM
2.1 BIM and Its Applications
2.2 The Need for Automation
3 Augmented, Virtual and Mixed Reality
3.1 Introduction to MR as the Emergence of AR and VR
3.2 AR, VR and MR in AEC
3.3 Integration of Gaming Engines in AEC
4 MR-Based Data Collection
4.1 Point Cloud Registration and Part Segmentation
4.2 Overlay and Augmentation
4.3 Data Production and Collection
4.4 Hybrid Gaming Engine and AR/VR Integration
4.5 Data-Set Creation and Management Systems
5 Results and Discussion
6 Conclusion
References
An Augmented Reality-Based Remote Collaboration Platform for Worker Assistance
1 Introduction
2 Augmented Reality for Training and Collaboration
3 System Design
4 Implementation
4.1 Spatial Mapping
4.2 Mapping 2D Coordinates in the 3D World
4.3 Real Time Communication
4.4 Keyframe Extraction
4.5 User Logs and Data Analytics
5 Experimental Results
5.1 Experiment Setup
5.2 Participants and Procedure
5.3 Results
6 Conclusion
References
Demand Flexibility Estimation Based on Habitual Behaviour and Motif Detection
1 Introduction
2 Datasets
3 Method
3.1 Routine Detection
3.2 Flexibility Detection
4 Results
5 Conclusion
References
Road Tracking in Semi-structured Environments Using Spatial Distribution of Lidar Data
1 Introduction
2 Related Work
3 Methodology
3.1 Data Pre-processing
3.2 Candidate Road Limits Detection
3.3 Road Tracking with SLAM
3.4 Final Trajectory Estimation
4 Experimental Evaluation
5 Conclusion
References
Image Segmentation of Bricks in Masonry Wall Using a Fusion of Machine Learning Algorithms
1 Introduction
2 Related Work
3 Proposed System
3.1 Hardware
3.2 Software
4 Experimental Evaluation
4.1 Dataset
4.2 Results
5 Conclusion and Discussion
References
Sentinel-2 and SPOT-7 Images in Machine Learning Frameworks for Super-Resolution
1 Introduction
2 Machine Learning Frameworks for Super-Resolution: Data, Procedures and Settings
3 Experimental Results
4 Conclusions
References
Salient Object Detection with Pretrained Deeplab and k-Means: Application to UAV-Captured Building Imagery
1 Introduction
2 Related Work
3 Semantic Segmentation with Deeplab
4 Segmentation Using Deep Features
5 Experiments
6 Conclusion
References
Clutter Slices Approach for Identification-on-the-Fly of Indoor Spaces
1 Introduction
2 Identification-on-the-Fly
2.1 Indoor Construction Spaces
2.2 Clutter-Slices
3 Clutter Slices Dataset
4 Clutter Slices Pipeline
5 Experiments and Results
5.1 Experimental Setup
5.2 Results
6 Conclusion
References
PRRS 2020 - 11th IAPR Workshop on Pattern Recognition in Remote Sensing
Preface
Organization
Program Committee Chairs
Program Committee
Remembering Both the Machine and the Crowd When Sampling Points: Active Learning for Semantic Segmentation of ALS Point Clouds
1 Introduction
2 Methodology
2.1 Employed Classifiers
2.2 Selection Strategies
2.3 Employed Oracle
2.4 Datasets
3 Results
3.1 Comparison of Selection Strategies
3.2 Comparison of Employed Classifiers
3.3 Comparison of Different Oracle Types
3.4 Estimation of Reachable Accuracies with Real Crowdworkers
4 Conclusion
References
Towards Urban Tree Recognition in Airborne Point Clouds with Deep 3D Single-Shot Detectors
1 Introduction
2 Methodology
2.1 Architecture
2.2 Training
2.3 Inference
2.4 Baseline
3 Data
4 Results
4.1 Training
4.2 Quantitative Assessment
4.3 Qualitative Assessment
4.4 Comparison with Semantic Segmentation
5 Conclusion and Outlook
References
Shared-Space Autoencoders with Randomized Skip Connections for Building Footprint Detection with Missing Views
1 Introduction
2 Related Work
3 Methodology
3.1 Problem Setting
3.2 Enforcing a Shared Space
3.3 Randomized Skip Connections
3.4 Loss Functions
4 Data and Data Preprocessing
4.1 Data
4.2 Data Preprocessing
5 Experiments and Results
5.1 Training
5.2 Performance Evaluation
6 Conclusions
References
Assessment of CNN-Based Methods for Poverty Estimation from Satellite Images
1 Introduction
2 Description of the Selected CNN-Based Methods
2.1 Nighttime Light
2.2 Land Use
2.3 Contrastive Spatial Analysis
2.4 Regression Step
2.5 Comparison Issues
3 A Common Framework to Assess the CNN-Based Methods
3.1 Framework Specifications
3.2 Data Description
3.3 Metrics and Evaluation Goal
3.4 Implementation
3.5 Results
4 How to Improve the Results?
4.1 Handling the Geographic Perturbation
4.2 Combining the Approaches
4.3 Experiments and Analysis
5 Conclusion and Future Work
References
Using a Binary Diffractive Optical Element to Increase the Imaging System Depth of Field in UAV Remote Sensing Tasks
1 Introduction
2 Radially Symmetric Binary Phase Apodization
3 Calculation of Binary Diffractive Optical Element
4 Detecting Images of House Numbers
5 Conclusions
References
Self-supervised Pre-training Enhances Change Detection in Sentinel-2 Imagery
1 Introduction
2 Methods
2.1 Change Detection Pipeline
2.2 Pretext Tasks for Self-supervision
3 Data and Setup
3.1 Datasets
3.2 Setup
4 Results and Discussion
5 Conclusions
References
Early and Late Fusion of Multiple Modalities in Sentinel Imagery and Social Media Retrieval
1 Introduction
2 Related Work
3 Methodology
3.1 Early Fusion in Satellite Image Retrieval
3.2 Late-Fusion Approach to Retrieve Relevant Social Media Content
4 Experiments
4.1 Datasets Description
4.2 Results
5 Conclusion
References
RISS 2020 - International Workshop on Research and Innovation for Secure Societies
Workshop on Research and Innovation
for Secure Societies (RISS)
General Chairs and Program Committee
Additional Reviewers
SURVANT: An Innovative Semantics-Based Surveillance Video Archives Investigation Assistant
1 Introduction
2 SURVANT in a Nutshell
2.1 The SURVANT Platform
2.2 Video Analysis for Object Tracking and Event Detecting
2.3 The Complex Query Formulator
2.4 The Indexer: Visual Similarity and People Re-identification
2.5 The Trajectory Miner
2.6 The Reasoner and the Semantic Repository
3 Use Cases
4 Results
4.1 Video Analysis for Object Tracking and Event Detecting
4.2 Event Reasoning
4.3 Trajectory Miner
5 Conclusions
References
Automatic Fake News Detection with Pre-trained Transformer Models
1 Introduction
2 Fake News
3 State-of-the-Art
3.1 Transformer and Language Models
3.2 Related Work
4 Methodology
4.1 Data Distribution
4.2 Preprocessing
5 Experiments
6 Results
7 Conclusion and Future Work
References
A Serverless Architecture for a Wearable Face Recognition Application
1 Introduction
2 Requirements
2.1 Overview of the AR Face Recognition System Components
2.2 Functional Requirements
2.3 Nonfunctional Requirements
3 AR Assistance Architecture
3.1 Serverless Architecture
3.2 Sequence Diagram
4 Experimental Setup
5 Conclusions
5.1 Further Developments
References
RGB-D Railway Platform Monitoring and Scene Understanding for Enhanced Passenger Safety
1 Introduction
2 Related State of the Art
3 Proposed Methodologies and Systemic Concept
3.1 3D Multi-object Detection and Tracking
3.2 2D Object Detection and Tracking Schemes
3.3 Fusion of MOT Results
4 The RailEye3D Railway Platform Dataset
5 Results and Discussion
6 Conclusion and Outlook
References
A Survey About the Cyberbullying Problem on Social Media by Using Machine Learning Approaches
1 Introduction
2 Related Works
3 Open Issues
4 Conclusions
References
Author Index