Proceedings of the Satellite Workshops of ICVGIP 2021

Description

This book constitutes peer-reviewed proceedings of satellite workshops of the 12th Indian Conference on Computer Vision, Graphics, and Image Processing (ICVGIP 2021). The book focuses on medical image processing, digital heritage, document analysis and recognition, and computer vision applications. The first part includes submissions on digital archiving and restoration methods with interesting and innovative research components. The second part focuses on medical imaging modalities including MRI, X-ray, CT, imaging in nuclear medicine, medical ultrasound, optical and confocal microscopy, and video and range data images. The third part deals with document analysis and recognition and focuses on text recognition, document layout analysis, understanding, historical and degraded document analysis, datasets, performance evaluation, metrics, etc. The fourth part of this book includes research work from academia and industry across the globe on smart, innovative, and practical applications of computer vision for industrial and societal impact. This book shares innovative ideas, experience and expertise, and ongoing research ideas and will be helpful for researchers and practitioners in academia and industry. Contents About the Editors Workshop on Digital Heritage (WDH) Systematic Approach to Tuning a Deep CNN Classifying Bharatanatyam Mudras 1 Introduction 2 Literature Review 3 Dataset Description 4 Methodology 4.1 Convolutional Neural Network 4.2 Hyperparameter Tuning 5 Results and Discussion 5.1 Architecture 5.2 Batch Sizes 5.3 Dropout Probability 5.4 Validation Split Percentage 5.5 Optimisers and Learning Rate 5.6 Summary 6 Conclusion References Comparative Analysis of Neural Architecture Search Methods for Classification of Cultural Heritage Sites 1 Introduction 2 Background Study 2.1 Search Space 2.2 Search Strategy 2.3 Evaluation Strategy 3 Methodology 4 Results and Discussion 5 Conclusion References Heritage Representation of Kashi Vishweshwar Temple at Kalabgoor, Telangana with Augmented Reality Application Using Photogrammetry 1 Introduction 1.1 Photogrammetry 1.2 Augmented Reality 1.3 Historical Background 2 Methodology 3 Results and Analysis 4 Conclusion References Augmented Data as an Auxiliary Plug-In Toward Categorization of Crowdsourced Heritage Data 1 Introduction 2 Related Works 3 Categorization of Crowdsourced Heritage Data 4 Experiments 4.1 Dataset 4.2 Training Setup 4.3 Evaluation Metrics 5 Results and Discussions 5.1 Ablation Study 6 Conclusions References Evolution of Bagbazar Street Through Visibility Graph Analysis (1746–2020) 1 Introduction 2 Space Syntax Method 3 Kolkata: Bag Bazar Street 4 Methodology 4.1 Data and Survey 4.2 Analysis Method 5 Result and Discussion 5.1 General Description 5.2 Axial Analysis 6 Conclusion References Mapping Archaeological Remains of 14th Century Fort of Jahanpanah Using Geospatial Analysis 1 Introduction 2 Methodology 2.1 Analyses of Geospatial Data 2.2 Geospatial Studies of Archaeological Sites 3 Conclusion References Spatial Analysis and 3d Mapping Historic Landscapes—Implications of Adopting an Integrated Approach in Simulation and Visualization of Landscapes 1 Introduction 2 Digitizing Heritage—A Brief Overview 3 Context 3.1 Badami and Its Immediate Environs 4 3D Landscape Model Generation 4.1 Building Model Generation of Sites 1 and 2 4.2 Building Model Generation of Site 3: 5 Concluding Remarks References Medical Image Processing (MedImage) HSADML: Hyper-Sphere Angular Deep Metric Based Learning for Brain Tumor Classification 1 Introduction 2 Methodology 2.1 SphereFace Loss 2.2 Backbone Architecture 2.3 Classification and Testing 2.4 Implementation and Network Training 3 Experiments, Results and Discussions. 3.1 Dataset and Experimental Protocol 3.2 Performance Metrics 3.3 Experimental Analysis 4 Conclusion and Future Aspect References Document Analysis and Recognition (DAR) Model Compression Based Lightweight Online Signature Verification Framework 1 Introduction 2 Literature Survey 3 Proposed Online Signature Verification Model 3.1 The Proposed Pruning Technique 4 Ablation Study of Various Pruned Models 4.1 Experimentation Setup 4.2 Ablation Study 5 Comparative Study 6 Conclusion and Future Work References End-to-End Transformer-Based Architecture for Text Recognition from Document Images 1 Introduction 2 Background 3 Related Work 4 Proposed Framework 5 Methodology 5.1 Super Resolution and Segmentation 5.2 Global Attention 5.3 Normalized Attention 5.4 Decoder 6 Experimental Results 6.1 Dataset 6.2 Results References A Hybrid Approach for Table Detection in Document Images 1 Introduction 2 Proposed Method 2.1 Dataset 2.2 Preprocessing and Feature Detection 2.3 Classification of Words 2.4 Machine Learning Approach 2.5 Deep Learning Approach 2.6 Hybrid Approach 3 Results and Discussions 4 Conclusion References Workshop on Computer Vision Applications (WCVA) The Ikshana Hypothesis of Human Scene Understanding 1 Introduction 2 Related Work 3 Method 3.1 Ikshana (the Eye) Hypothesis 3.2 IkshanaNet Architecture 4 Experiments 4.1 Experimental Setup 4.2 Experiments on Cityscapes 4.3 Experiments on Camvid 5 Validity Threats 6 Conclusion 6.1 Code References Worst-Case Adversarial Perturbation and Effect of Feature Normalization on Max-Margin Multi-label Classifiers 1 Introduction 2 Our Approach 2.1 Generating Adversarial Samples 2.2 Feature Normalization to Safeguard Against Adversarial Attack 3 Experiments and Discussion 3.1 Datasets 3.2 Evaluation Metrics 3.3 Benchmark Max-Margin Multi-label Methods 3.4 Results and Discussion 4 Summary and Conclusion References Catch Me if You Can: A Novel Task for Detection of Covert Geo-Locations (CGL) 1 Introduction 2 Related Work 3 Dataset Details 3.1 Images 3.2 Annotations 4 Adaptation of Existing Models for CGL Detection 4.1 YOLOv3 4.2 MobileNetv2 4.3 HRNetv2 5 Proposed Method 5.1 Notations 5.2 Encoder 5.3 Depth-Aware Feature Learning Block (DFLB)—Auxiliary Decoder 5.4 CGL Segmentation Block (CGLSB) 5.5 Geometric Transformation Equivariance (GTE) Loss 5.6 Intraclass Variance Reduction (IVR) Loss 5.7 Testing 6 Results and Experiments 6.1 Evaluation Metric 6.2 Quantitative Results 6.3 Qualitative Results 7 Conclusion and Discussion References MATIC: Memory-Guided Adaptive Transformer for Image Captioning 1 Introduction 2 Related Work 2.1 CNN-RNN Based Models 2.2 Transformer Based Models 3 Methodology 3.1 Overview 3.2 Visual Encoder 3.3 Linguistic Decoder 3.4 Training 4 Experiments and Results 4.1 Dataset 4.2 Training Details 4.3 Quantitative and Qualitative Results 5 Conclusion References Semantic Map Injected GAN Training for Image-to-Image Translation 1 Introduction 2 Proposed Semantic Map Injected GAN Training 3 Experimental Setup 3.1 GAN Models Used 3.2 Datasets Used 3.3 Metrics Used 4 Experimental Results and Analysis 5 Conclusion References TextGen3D: A Real-Time 3D-Mesh Generation with Intersecting Contours for Text 1 Introduction 2 Literature Review 3 Algorithm 3.1 Contour Generator 3.2 Contour Processor 3.3 Mesh Generator 4 Results 4.1 Dynamic Sampling 4.2 Intersection Removal 4.3 Mesh Quality 5 Conclusion References