Pattern Recognition: 6th Asian Conference, ACPR 2021, Jeju Island, South Korea, November 9–12, 2021, Revised Selected Papers, Part I (Lecture Notes in Computer Science, 13188)

This two-volume set LNCS 13188 - 13189 constitutes the refereed proceedings of the 6th Asian Conference on Pattern Recognition, ACPR 2021, held in Jeju Island, South Korea, in November 2021. The 85 full papers presented were carefully reviewed and selected from 154 submissions. The papers are organized in topics on: classification, action and video and motion, object detection and anomaly, segmentation, grouping and shape, face and body and biometrics, adversarial learning and networks, computational photography, learning theory and optimization, applications, medical and robotics, computer vision and robot vision. Preface Organization Contents – Part I Contents – Part II Classification One-Shot Image Learning Using Test-Time Augmentation 1 Introduction 2 Related Work 3 Problem Definition 4 Proposed Method 5 Experimental Methods 5.1 Datasets 5.2 Backbone 5.3 Training Methods 5.4 Experimental Conditions 6 Experimental Results 7 Conclusion References Offline Handwritten Mathematical Expression Recognition via Graph Reasoning Network 1 Introduction 2 Related Work 2.1 Grammar Based Methods 2.2 Image-to-Markup Methods 2.3 Graph Parsing Methods 3 Method 3.1 Overview 3.2 Symbol Localization 3.3 Graph Reasoning Network 3.4 Training Details 4 Experiment 4.1 Datasets 4.2 Results 4.3 Visualization 5 Conclusion References Circulant Tensor Graph Convolutional Network for Text Classification 1 Introduction 2 Related Work 3 Preliminaries 3.1 Tensor Notation 3.2 Graph Construction 4 The Proposed Method 4.1 Circulant Tensor Graph Convolutional Network 4.2 Heterogeneous Attention Module 4.3 Loss Function 5 Experiments 5.1 Dataset 5.2 Implementation Details 5.3 Comparison with State-of-the-art 5.4 Ablation Study 5.5 Tensor Graph Convolutional Network 6 Conclusion References NASformer: Neural Architecture Search for Vision Transformer 1 Introduction 2 Related Work 2.1 Vision Transformer 2.2 Neural Architecture Search 3 Preliminary 3.1 Window Based Self-Attention 3.2 One-shot NAS 4 NASformer 4.1 Dilated Window Based Self-Attention 4.2 Overall Architecture of NASformer 4.3 Search Space for Heterogeneous Transformer 4.4 Search Pipeline 5 Experiments 5.1 Implementation Details 5.2 Image Classification 5.3 Object Detection 5.4 Semantic Segmentation 5.5 Ablation Study 6 Conclusion References Interference Distillation for Underwater Fish Recognition 1 Introduction 2 Related Work 2.1 Underwater Fish Recognition 2.2 Knowledge Distillation 3 The Proposed Method 3.1 Supervised Training 3.2 Distillation by Feature Similarity Alignment 3.3 Distillation by KL-Divergence 4 Experiments 4.1 Training Setting 4.2 Results and Comparison 4.3 Ablation Study 4.4 Parameter Selection and Analysis 5 Conclusions References Action and Video and Motion Contact-Less Heart Rate Detection in Low Light Videos 1 Introduction 2 Related Works 3 Method 3.1 Artificial Illumination 3.2 Color Magnification 3.3 Face Detection and ROI Selection 3.4 Extraction of Frequency Spectrums from Color Variation Data 3.5 1D Convolutional Neural Network 3.6 Post-processing 4 Experimental Details 4.1 Dataset Description 4.2 Training 5 Results and Discussion 5.1 Evaluation Metrics 5.2 Performance on Low Light Videos 5.3 Performance on Normal Light Videos 5.4 Effect of Color Magnification 6 Conclusion References Spatio-temporal Weight of Active Region for Human Activity Recognition 1 Introduction 2 Proposed Method 2.1 Preprocessing 2.2 Body Joint Exploitation 2.3 Full-Body Image Representation 2.4 Spatio-temporal Weight for Classification 3 Experiment 4 Conclusion and Future Work References Motor Imagery Classification Based on CNN-GRU Network with Spatio-Temporal Feature Representation 1 Introduction 2 Method 2.1 Spatio-temporal Feature Generation 2.2 Classification with CNN-GRU Network 3 Data Description and Evaluation 3.1 BCI Competition IV_2a Data 3.2 Baseline Methods 4 Results 4.1 Comparison with Baseline Methods 4.2 Comparison Spatial Feature Representation Only 5 Discussion and Conclusion References Planar Motion Estimation for Multi-camera System 1 Introduction 2 Generalized Camera Model 3 Our Approach 3.1 Generalized Essential Matrix Under Planar Motion 3.2 The Linear 6-Point Algorithm 3.3 The Minimal 3-Point Algorithm 4 Experiments 4.1 Computational Complexity 4.2 Experiments on Synthetic Data 4.3 Experiments on Real Image Sequences 5 Conclusion References SARNN: A Spatiotemporal Prediction Model for Reducing Error Transmissions 1 Introduction 2 Related Work 3 Methods 3.1 Sep-LSTM 3.2 ST-Attention 3.3 SC-Block 3.4 SARNN 4 Experiments 4.1 Moving Mnist Dataset 4.2 KTH Action Dataset 4.3 Limitations of SARNN 5 Conclusions References 3OFRR-SLAM: Visual SLAM with 3D-Assisting Optical Flow and Refined-RANSAC 1 Introduction 2 Related Work 3 System Overview 4 Tracking 4.1 3D-Assisting Optical Flow Tracker 4.2 Refined-RANSAC and Camera Pose Estimation 4.3 Relocalization 5 Mapping 6 Experiment 6.1 Evaluation on TUM-RGBD Dataset 6.2 Evaluation of Refined-RANSAC 6.3 Application 7 Conclusions and Future Work References Motor Imagery Classification Based on Local Log Riemannian Distance Matrices Selected by Confusion Area Score 1 Introduction 2 Method 2.1 Local Region and Local Covariance Matrix 2.2 Review of Log Riemannian Distance Matrix 2.3 Local Region Selection 2.4 Forming Feature Vector 3 Data and Experiments 4 Result and Discussion 5 Conclusion References Distance-GCN for Action Recognition 1 Introduction 2 Distance Graph Convolutional Networks 2.1 Graph Construction 2.2 Graph Convolution 2.3 Subset Attention Mask Module 2.4 Full Network Architecture 3 Experiment 3.1 Dataset and Experiment Settings 3.2 Comparisons of Range Activation Functions 3.3 Ablation Study on Distance Graph and Attention Module 3.4 Comparisons with the State of the Arts 3.5 Visualization of Learnt Distance Ranges 4 Conclusion and Future Work References Causal Intervention Learning for Multi-person Pose Estimation 1 Introduction 2 Related Work 3 CIposeNet Based Multi-Person Pose Estimation 3.1 Overview 3.2 Causal Assumption of Keypoints 3.3 Causal Intervention via Backdoor Adjustment for Keypoints 4 Experiments 4.1 Dataset 4.2 Ablation Study 4.3 The Results Analysis of CIposeNet 5 Conclusion References Object Detection and Anomaly Weakly Supervised Salient Object Detection with Box Annotation 1 Introduction 2 Related Work 2.1 Weakly-supervised Salient Object Detection 2.2 Box-supervised Pixel-wise Methods 3 Approach 3.1 Vector Supervision 3.2 Point Supervision 3.3 Plane Supervision 3.4 Training with Multiple Supervisions 4 Experiments 4.1 Setup 4.2 Implementation Details 4.3 Comparisons with the State-of-the-Art 4.4 Ablation Study 5 Conclusions References Foreground-Background Collaboration Network for Salient Object Detection 1 Introduction 2 Related Work 3 Method 3.1 Overview of Network Architecture 3.2 Attention-Aware Foreground-Background Separation Module 3.3 Spatial Feature Refinement Module 4 Experiments 4.1 Datasets and Evaluation Metrics 4.2 Implementation Details 4.3 Comparison with State-of-the-arts 4.4 Ablation Studies 5 Conclusion References Attention Guided Multi-level Feedback Network for Camouflage Object Detection 1 Introduction 2 Related Work 2.1 Generic Object Detection 2.2 Camouflage Object Detection 2.3 Attention Mechanism 3 The Proposed Method 3.1 Attention Guided Capture Module 3.2 Multi-level Feedback Module 4 Experiment 4.1 Datasets and Evaluation Criteria 4.2 Implementation Details 4.3 Loss Function 4.4 Comparisons with State-of-the Arts 4.5 Ablations 5 Conclusion References Confidence-Aware Anomaly Detection in Human Actions 1 Introduction 2 Related Work 3 Proposed Model 3.1 Confidence-weighted Graph Convolution Network 3.2 Temporal Confident Weighted Graph Convolution 3.3 Network Architecture 3.4 Loss Function 3.5 Anomaly Detection 4 Experiments 4.1 Data Preparation 4.2 Implementation Details 4.3 Results and Discussion 4.4 Ablation Study 5 Conclusion References Segmentation, Grouping and Shape 1D Self-Attention Network for Point Cloud Semantic Segmentation Using Omnidirectional LiDAR 1 Introduction 2 Related Work 2.1 Voxel-Based Method 2.2 Image-Based Method 2.3 3D Point Cloud-Based Method 2.4 1D-CNN for Pedestrian Detection 3 Proposed Method 3.1 Network Structure 3.2 Preprocessing Point Cloud Data 3.3 Normalization of Intensity Value 3.4 1 Dimensional Self-Attention Block (1D-SAB) 4 Experiments 4.1 Experimental Summary 4.2 Evaluating the Effectiveness of Intensity 4.3 Evaluation the Effectiveness of 1D-SAB 4.4 Qualitative Evaluation 4.5 Comparison of Accuracy with Other Methods 4.6 Comparison of Processing Speed 5 Conclusion References COMatchNet: Co-Attention Matching Network for Video Object Segmentation 1 Introduction 2 Related Work 3 Method 3.1 Network Overview 3.2 Co-Attention 3.3 Matching 4 Experiments 4.1 Implement Details 4.2 Comparing to Other Advanced 4.3 Ablation Study 4.4 Qualitative Results 5 Conclusion References Unsupervised Domain Adaptive Point Cloud Semantic Segmentation 1 Introduction 2 Related Work 2.1 Unsupervised Domain Adaptation 2.2 Deep Point Cloud Semantic Segmentation 3 Method 3.1 Overview 3.2 Multi-level Feature Consistency 3.3 Feature Bank Based Cycle Association 3.4 Training 4 Experiments 4.1 Implementation Details 4.2 Datasets 4.3 Performance Comparison 4.4 Ablation Study 5 Conclusion References Towards the Target: Self-regularized Progressive Learning for Unsupervised Domain Adaptation on Semantic Segmentation 1 Introduction 2 Related Work 3 Proposed Method 3.1 Baseline for Cross-Domain Adaptation 3.2 Source Label Relaxation for Cross-Domain Adaptation 3.3 Dual-Level Self-regularization for Within-Domain Adaptation 4 Experiment 4.1 Datasets and Evaluation Metrics 4.2 Implementation Details 4.3 Comparisons 4.4 Ablation Study 5 Conclusion References TFNet: Transformer Fusion Network for Ultrasound Image Segmentation 1 Introduction 2 Method 2.1 Transformer Fuse Module 2.2 Loss Function 3 Experiments 3.1 Datasets 3.2 Experimental Settings 3.3 Results 3.4 Analysis 4 Conclusion References Spatial Pyramid-based Wavelet Embedding Deep Convolutional Neural Network for Semantic Segmentation 1 Introduction 2 Related Work 2.1 Semantic Segmentation 2.2 Wavelet Transform 2.3 Multi-scale Fusion 3 Spatial Pyramid-based Wavelet (SPW) Embedding Deep Convolutional Structure 3.1 WTD and IWTU 3.2 Spatial Pyramid Structure 4 Experiment 4.1 Implementation Details 4.2 Ablation Experiment 4.3 Comparison with the State-Of The-Arts 5 Conclusion References CFFNet: Cross-scale Feature Fusion Network for Real-Time Semantic Segmentation 1 Introduction 2 Related Work 3 Methodology 3.1 Architecture 3.2 Lightweight Residual Block 3.3 Cross-scale Feature Fusion Module 4 Experiments 4.1 Datasets 4.2 Implementation Details 4.3 Parameters and FLOPs 4.4 Accuracy and Speed 4.5 Compare of Residual Blocks 4.6 Ablation Study 5 Conclusion References Semantic Segmentation and Depth Estimation with RGB and DVS Sensor Fusion for Multi-view Driving Perception 1 Introduction 2 Related Work 2.1 Deep Multi-task Learning 2.2 Sensor Fusion in Deep Learning 3 Methodology 3.1 Dataset and Input/Output Representation 3.2 Network Architecture 3.3 Experiment Setup 4 Result and Discussion 4.1 Performance Gain by Sensor Fusion 4.2 Comparison with Another Model 5 Conclusion and Future Work References ARTSeg: Employing Attention for Thermal Images Semantic Segmentation 1 Introduction 2 Related Work 3 Methods 4 Experimentation and Results 4.1 Thermal Semantic Segmentation Dataset 4.2 Training Details 4.3 Evaluation Metrics 4.4 Results 5 Conclusion References Robust Frequency-Aware Instance Segmentation for Serial Tissue Sections 1 Introduction 2 Related Work 2.1 Instance Segmentation 2.2 Channel Attention Mechanism 3 Method 3.1 DCT Frequency on Feature Map 3.2 FANet 4 Experiments 4.1 Dataset 4.2 Implementation Details 4.3 the Specific Frequency Components 4.4 Instance Segmentation on the Benchmark 5 Conclusions References Joint Semantic Segmentation and Edge Detection for Waterline Extraction 1 Introduction 2 Related Work 3 Network Architecture 3.1 Detail Branch 3.2 Semantic Branch 3.3 Aggregation Module 3.4 Loss Function 4 Experiments 4.1 Dataset 4.2 Train Setting 4.3 Evaluation Metric 4.4 Results 5 Conclusions References Face and Body and Biometrics Comparing Facial Expression Recognition in Humans and Machines: Using CAM, GradCAM, and Extremal Perturbation 1 Introduction 2 Related Work 2.1 Automatic FER 2.2 Human FER 2.3 Transfer Learning 3 Dataset 4 Human Experiment 4.1 Participants 4.2 Methods and Task 4.3 Results 5 Algorithms 5.1 Model Training 5.2 Model Results 5.3 Comparing Humans and Models 6 Conclusion and Future Work References Pose Sequence Generation with a GCN and an Initial Pose Generator 1 Introduction 2 Related Work 2.1 Image Generation 2.2 Video Generation 2.3 Pose Sequence Generation 2.4 Graph Convolution Network 3 Method 3.1 Proposed Model 3.2 Loss Functions 3.3 Pose Representation 4 Experiments 4.1 Dataset 4.2 Qualitative Evaluation 4.3 Interpolation of Latent Space 4.4 Quantitative Evaluation 4.5 Ablation Study 5 Conclusion References Domain Generalization with Pseudo-Domain Label for Face Anti-spoofing 1 Introduction 2 Related Work 2.1 DG Based Face Anti-spoofing Methods 2.2 Meta Learning for Domain Generalization (MLDG) 3 Method 3.1 Overview 3.2 Splitting Domains 3.3 Meta Learning for Domain Generalization 4 Experimental Evaluation 4.1 Experimental Settings 4.2 Experimental Comparison 5 Conclusion References Face Anti-spoofing via Robust Auxiliary Estimation and Discriminative Feature Learning 1 Introduction 2 Related Work 3 Proposed Method 3.1 Auxiliary Information 3.2 Network Architecture 3.3 Auxiliary Estimation and Discriminative Feature Learning 3.4 Live/Spoof Classification 4 Experiments 4.1 Experimental Setting 4.2 Ablation Study 4.3 Intra Testing and Cross Testing 5 Conclusion References PalmNet: A Robust Palmprint Minutiae Extraction Network 1 Introduction 2 Related Work 3 PalmNet 3.1 Gabor Amplitude-Phase Model 3.2 Backbone Network 3.3 Head Network 4 Experiment 4.1 Database 4.2 Implementation Details and Hyper-parameters 4.3 Evaluation 4.4 Conclusion and Future Work References Multi-stage Domain Adaptation for Subretinal Fluid Classification in Cross-device OCT Images 1 Introduction 2 Related Work 3 Method 3.1 Task-Independent Feature Alignment Module 3.2 Task-Specific Feature Alignment Module 4 Experiments 4.1 Materials 4.2 Experiment Protocols 4.3 Performance Evaluation 4.4 Results and Discussion 4.5 Ablation Study 5 Conclusion References SaME: Sharpness-aware Matching Ensemble for Robust Palmprint Recognition 1 Introduction 2 Related Work 3 Method 3.1 Framework of SaME 3.2 Multifeature Extraction 3.3 Sharpness-Based Matching Score 3.4 Matching Ensemble 3.5 ROI Augmentation Strategy 4 Experimental Results 4.1 Sharpness-Based Quality Indicator 4.2 Recognition Performance of the Proposed SaME 5 Conclusion References Weakly Supervised Interaction Discovery Network for Image Sentiment Analysis 1 Introduction 2 Related Work 2.1 Visual Sentiment Prediction 2.2 Weakly Supervised Detection 2.3 Graph Convolutional Network 3 Method 3.1 Sentiment Map Detection 3.2 Sentiment Interaction Extraction 3.3 Joint Training Strategy 4 Experiment 4.1 Datasets 4.2 Baselines 4.3 Implementation Details 4.4 Classification Results 4.5 Ablation Study 4.6 Detection Results 5 Conclusion References Joint Multi-feature Learning for Facial Age Estimation 1 Introduction 2 Related Work 2.1 Multi-feature Age Estimation 2.2 Age Estimator 3 Proposed Method 3.1 The Framework of the Proposed Method 3.2 Multiple Type Aging-Related Features Learning 3.3 Regression-Ranking Age Estimator 4 Experiments 4.1 Datasets and Preprocessing 4.2 Comparisons with the State-of-the-art 4.3 Ablation Analysis 5 Conclusion References Is CNN Really Looking at Your Face? 1 Introduction 2 Related Work 2.1 Face Recognition 2.2 Visualization of CNNs 2.3 Face Parsing 3 Datasets 3.1 Datasets for Face Parsing 3.2 Datasets for Face Recognition 4 Face Parsing for Face Recognition 5 Face Recognition with Face Parsing 6 Experiments and Discussion 6.1 Experimental Settings 6.2 Experimental Results 6.3 Discussion on Facial Parts 6.4 Discussion on Facial Features 7 Application to Visualization 8 Conclusion References NFW: Towards National and Individual Fairness in Face Recognition 1 Introduction 2 Related Work 3 NFW: National Face In-the-World 3.1 Image Collection 3.2 Image Filtering 3.3 Identity Selection 3.4 Pair Selection 4 Experiments on NFW 4.1 Results on National Bias 4.2 Results on Individual Bias 5 Conclusions and Future Work References Adversarial Learning and Networks Unsupervised Fabric Defect Detection Based on DCGAN with Component-Encoder 1 Introduction 2 Proposed Methods 2.1 Framework Description 2.2 Training Objective 3 Experiments 3.1 Datasets 3.2 Defect Detection 3.3 Evaluation Metrics 3.4 Implementation Details 3.5 Comparison with Other Methods 4 Conclusion References Proactive Student Persistence Prediction in MOOCs via Multi-domain Adversarial Learning 1 Introduction 2 Related Works 3 Proposed Method 3.1 Feature Extraction 3.2 Feature Representation 3.3 Multi-domain Adversarial Feature Representation (mDAFR) Learning 4 Experiments 4.1 Dataset 4.2 Implementation Details 4.3 Results and Comparative Study 5 Conclusion References Author Index