Image Analysis: 22nd Scandinavian Conference, SCIA 2023, Sirkka, Finland, April 18–21, 2023, Proceedings, Part I. (Lecture Notes in Computer Science)

This two-volume set (LNCS 13885-13886) constitutes the refereed proceedings of the 23rd Scandinavian Conference on Image Analysis, SCIA 2023, held in Lapland, Finland, in April 2023. The 67 revised papers presented were carefully reviewed and selected from 108 submissions. The contributions are structured in topical sections on datasets and evaluation; action and behaviour recognition; image and video processing, analysis, and understanding; detection, recognition, classification, and localization in 2D and/or 3D; machine learning and deep learning; segmentation, grouping, and shape; vision for robotics and autonomous vehicles; biometrics, faces, body gestures and pose; 3D vision from multiview and other sensors; vision applications and systems. Preface Organization Contents – Part I Contents – Part II Datasets and Evaluation LiDAR Place Recognition Evaluation with the Oxford Radar RobotCar Dataset Revised*-12pt 1 Introduction 2 Related Work 3 Oxford Radar RobotCar Place Recognition Benchmark 3.1 Overview 3.2 Sensors 3.3 Gallery, Training and Test Sequences 3.4 Evaluation Metrics 4 Baseline Method 5 Preliminary Experiments 6 Revised Location Ground Truth 7 Experiments 8 Conclusion References BrackishMOT: The Brackish Multi-Object Tracking Dataset*-12pt 1 Introduction 2 Related Work 2.1 Underwater MOT Datasets 2.2 Underwater Trackers 2.3 Synthetic Underwater Data 3 The BrackishMOT Dataset 3.1 Dataset Overview 3.2 BrackishMOT Splits 4 Synthetic Data Framework 5 Experiments 5.1 Qualitative Evaluation 6 Conclusion References Camera Calibration Without Camera Access - A Robust Validation Technique for Extended PnP Methods 1 Introduction 1.1 Background Motivation 1.2 Ethical Consideration 2 Related Work 3 Method 3.1 Motivating Example 3.2 Camera Calibration 3.3 Residual Error Model 3.4 Noise Estimation 3.5 Hypothesis Testing 4 Experiments 4.1 Synthetic Data 4.2 Lidar Measurements 4.3 Structure-from-Motion 5 Conclusion References CHAD: Charlotte Anomaly Dataset 1 Introduction 2 Related Work 3 Data Collection and Setup 4 Annotation Methodology 4.1 Anomaly Annotations 4.2 Person Annotations 4.3 Annotation Smoothing 5 CHAD Statistics 6 Metrics and Measurements 6.1 Receiver Operating Characteristic Curve 6.2 Precision-Recall Curve 6.3 Equal Error Rate 7 Evaluation 7.1 Standard Validation 7.2 Cross Validation 8 Conclusion References iDFD: A Dataset Annotated for Depth and Defocus 1 Introduction 2 Related Work 2.1 RGB-D Datasets 2.2 Datasets for Image Deblurring 2.3 Datasets for Joint DFD and Image Deblurring 3 iDFD Dataset for DFD and Image Deblurring 3.1 Data Acquisition 3.2 Data Preprocessing 4 Experimental Results 5 Conclusion References TBPos: Dataset for Large-Scale Precision Visual Localization*-12pt 1 Introduction 2 Related Work 2.1 Datasets 2.2 Algorithms 3 Visual Data Acquisition Procedure 4 Synthesizing Query Images 5 Experimental Results 5.1 Analysis of Pose Estimation Failure Cases 6 Conclusions References FinnWoodlands Dataset 1 Introduction 2 Related Works 3 Dataset Features 4 Data Collection 5 Experiments 5.1 Evaluation 5.2 Results 6 Conclusion References Re-identification of Saimaa Ringed Seals from Image Sequences 1 Introduction 1.1 Related Work 2 Proposed Method 2.1 Pipeline 2.2 Data Preprocessing 2.3 Feature Aggregation 2.4 Re-identification 3 Experiments 3.1 Data 3.2 Description of Experiments 3.3 Evaluation Criteria 3.4 Results 4 Conclusion References Action and Behaviour Recognition Attention-guided Boundary Refinement on Anchor-free Temporal Action Detection 1 Introduction 2 Related Work 2.1 Video Action Detection 2.2 Feature Extraction in Video Action Analysis 3 Proposed Method 3.1 Anchor-free Action Detection 3.2 Attention-guided Refinement Module 3.3 Temporal Attention Unit 4 Experimentation 4.1 Datasets and Settings 4.2 Ablation Study 4.3 Compare with State-of-the-Art 5 Conclusions References Spatio-temporal Attention Graph Convolutions for Skeleton-based Action Recognition*-12pt 1 Introduction 2 Related Works 2.1 Deep Learning on Graphs 2.2 Skeleton-based Action Recognition 3 Proposed Method 3.1 Preliminaries 3.2 Model Architecture 3.3 Spatial Modeling 3.4 Temporal Modeling 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Ablation Study 4.4 Results 5 Conclusion References Image and Video Processing, Analysis, and Understanding RELIEF: Joint Low-Light Image Enhancement and Super-Resolution with Transformers 1 Introduction 2 Background 2.1 Low-Light Image Enhancement 2.2 Image Super-Resolution 2.3 Vision Transformer 3 Method 3.1 Overall Pipeline 3.2 ECSWin Self-attention Transformer Block 3.3 Locally-Enhanced Feed-Forward Network 3.4 Locally-Enhanced Positional Encoding 4 Experiments and Analysis 4.1 Datasets 4.2 Evaluation Metrics 4.3 Implementation Details 4.4 Comparison with Existing Methods 4.5 Results 4.6 Ablation Studies 5 Conclusion References To Quantify an Image Relevance Relative to a Target 3D Object 1 Introduction 2 Deterministic Method for Relevance Evaluation 2.1 Salient Points in 3D and in 2D 2.2 Intersection and Union Maps 3 Relevance Score 4 Confidence Score 5 Validation Protocol 6 Results on Reference Rankings 7 Results on Real Images 8 Conclusion References Deep Active Learning for Glioblastoma Quantification 1 Introduction 2 Deep Active Learning for Glioma Segmentation 3 Experimental Setup and Results 4 Conclusion References Improved Sensitivity of No-Reference Image Visual Quality Metrics to the Presence of Noise 1 Introduction 2 Verification of METRIC’S Sensitivity to the Presence of Noise 2.1 Databases with MOS for Verification of NR-IQA Metrics 2.2 Dataset and Quality Criterion for Verification of Sensitivity of NR-IQA Metrics to the Presence of Noise 3 Comparative Analysis of Blind Noise Level Estimators 4 Combining Quality Prediction and Blind Noise Level Estimation 5 Conclusions References Rethinking Matching-Based Few-Shot Action Recognition 1 Introduction 2 Related Work 3 Method 3.1 Preliminaries 3.2 A Common Setup for Classifier and Matching-Based Approaches 3.3 Chamfer++ 4 Experiments 4.1 Implementation Details 4.2 Results 4.3 Chamfer Matching Ablation and Interpretability 5 Conclusion A Appendix A.1 Baseline Matching Functions A.2 Additional Ablations and Impact of Hyper-parameters References Accuracy of Parallel Distance Mapping Algorithms When Applied to Sub-Pixel Precision Transform 1 Introduction 1.1 EDT 1.2 AAEDT 1.3 This Study 2 Background 2.1 AAEDT 2.2 SKW 2.3 JFA 2.4 PBA 2.5 Problems with PBA for AAEDT 3 Method 3.1 Implementation 3.2 Test on Uniform Shapes 3.3 Test on Non-Uniform Shapes 4 Results 4.1 Test on Uniform Shapes 4.2 Test on Non-Uniform Shapes 5 Discussion 5.1 Test on Uniform Shapes 5.2 Test on Non-Uniform Shapes 6 Conclusions 7 Future Work References Distortion-Based Transparency Detection Using Deep Learning on a Novel Synthetic Image Dataset 1 Introduction 2 Related Work 3 Transparency Perception User Study 3.1 Experiment Design 3.2 Stimuli and Data Generation 3.3 Results 4 Artificial Transparency Detection Method 4.1 DISTOPIA Dataset 4.2 Classification 4.3 Evaluation 5 Conclusion and Future Work References Regenerated Image Texture Features for COVID-19 Detection in Lung Images 1 Introduction 2 Literature Survey 3 Proposed Methodology 3.1 Extraction of Texture Features 3.2 Feature Regeneration 4 Experiments and Results 4.1 Parameters of ANN Architecture 4.2 CT-scan Images 4.3 X-ray Images 4.4 Discussion of Results 5 Conclusions References Depth-Aware Image Compositing Model for Parallax Camera Motion Blur 1 Introduction 2 Related Work 3 Geometry of Camera Motion Blur 3.1 Fundamentals 3.2 In-plane Camera Motion 4 Image Compositing Blur (ICB) Model 4.1 Depth-Dependent Regions 4.2 Blur Kernels Synthesis 5 Neural Representations from Blur 6 Experiments 6.1 Evaluation Datasets 6.2 Model Validation 6.3 Neural Representations from Blur 7 Conclusion References Detection, Recognition, Classification, and Localization in 2D and/or 3D Affine Moment Invariants of Tensor Fields 1 Introduction 2 Literature Survey 3 Affine Tensor Field Moment Invariants 3.1 Covariant and Contravariant Indices 3.2 Contraction 3.3 Transformations of Tensor Fields 3.4 Moment Tensors 3.5 Construction of the Invariants 3.6 Tensor Field Affine Moment Invariants and Quadri-Layer Hypergraphs 4 Numerical Experiment 4.1 Obtaining Data and Visualization 4.2 Invariance 4.3 Template Matching 5 Conclusion References Fashion CUT: Unsupervised Domain Adaptation for Visual Pattern Classification in Clothes Using Synthetic Data and Pseudo-labels 1 Introduction 2 Related Work 3 Fashion CUT 4 Experiments 4.1 Zalando SDG Dataset 4.2 Evaluation on Zalando SDG Dataset 4.3 Synthetic Dataset Size 5 Conclusions References Long Range Object-Level Monocular Depth Estimation for UAVs 1 Introduction 2 Related Work 3 Methodology 3.1 YOLOX – Base Framework for 2D Object Detection 3.2 Depth Regression 3.3 Depth Bin Classification 3.4 Fitness Score 4 Experiments 4.1 Dataset 4.2 Experimental Setup 4.3 Results 5 Conclusion References RadarFormer: Lightweight and Accurate Real-Time Radar Object Detection Model 1 Introduction 2 Object Detection on Radar Data 3 The Proposed RadarFormer 3.1 Transformers 3.2 Attention Variation 3.3 Use of 2D Information 3.4 Effect of Receptive Fields and Residuals 4 Experiments 4.1 Baselines 4.2 Quantitative Results 4.3 Qualitative Results 4.4 Ablation Studies 5 Conclusion References Drawing and Analysis of Bounding Boxes for Object Detection with Anchor-Based Models 1 Introduction 1.1 Common Object Detection Framework with DL 2 Related Work 3 Experimentation 3.1 Datasets Used 3.2 Configuration 4 Results 5 Conclusion 6 Appendix References Raw or Cooked? Object Detection on RAW Images 1 Introduction 2 Related Work 3 Method 3.1 Downsampling RAW Images 3.2 Learnable ISP Operations 3.3 Our Raw Object Detector 4 Experiments 4.1 Dataset 4.2 Implementation Details 4.3 Quantitative Results 4.4 Qualitative Results 4.5 Parameter Evolution 5 Conclusion References Local Neighborhood Features for 3D Classification 1 Introduction 2 Related Works 3 Method 3.1 Neighborhood Point Distance 3.2 Directional Vectors as Neighborhood Features 3.3 Weights-Averaging of Checkpoints 4 Experiments 5 Experimental Results 6 Ablation Study 6.1 Distance vs. r-normalized Distance 6.2 Weight Averaging (Same Training Session) 6.3 Cost of Additional Neighborhood Features 7 Conclusion References 3D Point Cloud Registration for GNSS-denied Aerial Localization over Forests 1 Introduction 2 Related Work 3 Method 3.1 Registration Pipeline 3.2 SHOT-N 4 Evaluation 4.1 Test Areas 4.2 Datasets 4.3 Evaluation Metrics 5 Results 6 Summary and Discussion 7 Future Work References Cleaner Categories Improve Object Detection and Visual-Textual Grounding 1 Introduction 2 Related Work 2.1 Bottom-Up for Object Detection 2.2 Noisy Label Sets 3 Recap: Bottom-Up Faster R-CNN 4 Cleaning the Visual Genome Category Set 5 Experimental Setup 5.1 Datasets and Evaluation Metrics 5.2 Random Baseline 5.3 Implementation Details 6 Experiments 6.1 Object Detection 6.2 Feature Space Analysis 6.3 Visual Grounding Results 7 Conclusion and Future Work References Correction to: Image Analysis Correction to: R. Gade et al. (Eds.): SCIA 2023, LNCS 13885, https://doi.org/10.1007/978-3-031-31435-3 Author Index