Computer Vision and Image Processing : 7th International Conference, CVIP 2022, Nagpur, India, November 4–6, 2022, Revised Selected Papers, Part I

This two volume set (CCIS 1776-1777) constitutes the refereed proceedings of the 7th International Conference on Computer Vision and Image Processing, CVIP 2022, held in Nagpur, India, November 4–6, 2022. The 110 full papers and 11 short papers were carefully reviewed and selected from 307 submissions. Out of 121 papers, 109 papers are included in this book. The topical scope of the two-volume set focuses on Medical Image Analysis, Image/ Video Processing for Autonomous Vehicles, Activity Detection/ Recognition, Human Computer Interaction, Segmentation and Shape Representation, Motion and Tracking, Image/ Video Scene Understanding, Image/Video Retrieval, Remote Sensing, Hyperspectral Image Processing, Face, Iris, Emotion, Sign Language and Gesture Recognition, etc. Preface Organization Contents – Part I Contents – Part II Anomaly Detection in ATM Vestibules Using Three-Stream Deep Learning Approach 1 Introduction 2 Related Work 3 Dataset 3.1 Previous Datasets 3.2 Data Collection 3.3 ATM-I Dataset 3.4 ATMA-V Dataset 4 Proposed Methodology 4.1 Contextual Feature Extraction 4.2 Motion Feature Extraction 4.3 Spatial Feature Extraction 4.4 Classification Model 4.5 Anomaly Score Function 5 Experiment and Results 5.1 Implementation Details 5.2 Datasets 5.3 Ablation Study 5.4 Quantitative Comparison with Previous Works 5.5 Qualitative Results 6 Conclusion and Future Work References MIS-Net: A Deep Residual Network Based on Memorised Pooling Indices for Medical Image Segmentation 1 Introduction 2 Related Works 3 Proposed MIS-Net Model 3.1 Training or Network Parameters 4 Experimental Setup 4.1 Dataset Summary 4.2 Performance Analysis Metrics 4.3 Hyperparameter Settings 5 Experimental Results and Discussion 5.1 Liver Segmentation Using the LITS and 3DIRCADb Datasets 5.2 Lung Segmentation Using the COVID-19 CT Dataset 5.3 Covid-19 Infection Segmentation Using the COVID-19 CT Dataset 6 Conclusion References HD-VAE-GAN: Hiding Data with Variational Autoencoder Generative Adversarial Networks 1 Introduction 2 Related Work 3 Proposed Method 3.1 Problem Formulation 3.2 VAE Embedder Network 3.3 Extractor Network 3.4 Discriminator Networks 3.5 Training 3.6 Transmission Attacks 4 Experiment 4.1 Dataset Details 4.2 Implementation Details 4.3 Results 5 Analysis 5.1 Feeding an Empty Container Image to the Extractor 5.2 Pixel Level Distortion Due to Embedding 5.3 What if an Attacker Gets Access to the Cover Image 6 Conclusion References Share-GAN: A Novel Shared Task Training in Generative Adversarial Networks for Data Hiding 1 Introduction 2 Related Work 3 Proposed Method 3.1 Problem Formulation 3.2 Generator Network 3.3 Discriminator Network 3.4 Training GAN for Message Embedding 3.5 Training on Attack Simulations 3.6 Implementation Details 4 Experiment 4.1 Dataset Details 4.2 Metrics Used 4.3 Grayscale Messaging 4.4 Binary Messaging 5 Analysis 5.1 Feeding a Non-encoded Image to the Extractor 5.2 What if an Attacker Gets Access to the Cover Image 6 Conclusion References Hiding Video in Images: Harnessing Adversarial Learning on Deep 3D-Spatio-Temporal Convolutional Neural Networks 1 Introduction 2 Related Work 3 Proposed Method 3.1 Problem Formulation 3.2 Embedder Network 3.3 Extractor Network 3.4 Embedder Discriminator Network 3.5 Video Discriminator Networks 3.6 Training Procedure for Vid-in-Img-RAN 3.7 Training Procedure for Vid-in-Img-3D-GAN 4 Experiment 4.1 Dataset Details 4.2 Results 5 Conclusion References An Explainable Transfer Learning Based Approach for Detecting Face Mask 1 Introduction 2 Related Works 3 Proposed Model 3.1 Dataset and Preprocessing 3.2 Classifier 4 Result and Discussion 4.1 Experiment Setup 4.2 Model Comparison 4.3 Explainability of the Model 4.4 Performance of Model on Distortions 4.5 Real-Time Implementation of the Model 5 Conclusion References Left Ventricle Segmentation of 2D Echocardiography Using Deep Learning 1 Introduction 2 Contribution 3 Methodology 3.1 U-Net 3.2 Res U-Net 3.3 Res34 U-Net 3.4 Vgg16 U-Net 4 Experimental Results 4.1 Dataset 4.2 Network Training 4.3 Evaluation Metrics 4.4 Segmentation Results 5 Conclusion References Multi Modal 2-D Canvas Based Gallery Content Retrieval 1 Introduction 2 Related Work 2.1 Text Based Image Retrieval 2.2 Sketch Based Image Retrieval 2.3 Handwriting Recognition 3 Proposed Approach 3.1 Sketch Segmentation and Classification 3.2 Handwritten Text Recognition and Synonym Matching 3.3 Sketch Recognition 3.4 Finer Feature Classifier 3.5 Positional Content Retrieval 4 Experiments and Results 4.1 Training Data-Sets 4.2 Testing Environment 4.3 Experiments 4.4 User Survey 5 Conclusion and Future Work References A Segmentation Based Robust Fractional Variational Model for Motion Estimation 1 Introduction 2 Contribution 3 Mathematical Formulation of the Proposed Model 3.1 Level Set Segmentation Framework Based Fractional Order Variational Model 3.2 Charbonnier Norm Based Segmented Data Penalty Term 3.3 Segmented Smoothness Constraint 4 Experiments, Results and Discussions 4.1 Datasets 4.2 Performance Measure 4.3 Experimental Discussions 5 Conclusions and Future Work References FlashGAN: Generating Ambient Images from Flash Photographs 1 Introduction 2 Related Work 3 Methodology 3.1 Dataset 3.2 Conditional GAN 3.3 Training 4 Results 5 Conclusion and Future Scope References CT Image Synthesis from MR Image Using Edge-Aware Generative Adversarial Network 1 Introduction 2 Related Works 3 Methodology 3.1 Data Acquisition and Preprocessing 3.2 EaGAN for MRI-to-CT Synthesis 3.3 Training 4 Results 5 Conclusion References Modified Scaled-YOLOv4: Soccer Player and Ball Detection for Real Time Implementation 1 Introduction 2 Literature 3 Proposed Methodology 4 Dataset and Experimentation 5 Real-Time Implementation and Results 5.1 Real-Time Testing Results on Jetson TX2 6 Conclusion and Future Scope References CandidNet: A Novel Framework for Candid Moments Detection 1 Introduction 2 Related Work 3 Model Architecture 4 Feature Extraction Stage 5 Classification Stage 6 Scoring Stage 7 Experimentatal Setup 7.1 Datasets 7.2 Training Details 8 Ablation Studies 8.1 CandidNet Modules 8.2 Feature Extraction Stage 8.3 Classification Stage Before Scoring Stage 9 Results and Discussion 10 Conclusion References Cost Efficient Defect Detection in Bangle Industry Using Transfer Learning 1 Introduction 1.1 Problem Scope 1.2 Contributions 2 Related Work 3 Dataset 4 Methods 4.1 Traditional Methods 4.2 Proposed Method 4.3 Pre-processing 4.4 Feature Extraction 4.5 Classification 5 Experiments and Results 5.1 Binary Classification 5.2 Combining Traditional Features 5.3 Variations in Pre-processing Techniques 5.4 Extending to Detect Size Defects 6 Conclusion 7 Ethical Impact References Single Image Dehazing Using Multipath Networks Based on Chain of U-Nets 1 Introduction 2 Related Work 3 Method 3.1 Motivation Behind the New Architecture 3.2 Ladder-Net and U-Net 3.3 Advantages of Using Dense Connections 3.4 Computational Complexity 3.5 Loss Function 4 Experiments 4.1 Dataset Setup 4.2 Training Details 4.3 Metrics Used 4.4 Compared Methods 4.5 Quantitative Metrics 4.6 Qualitative Comparison 5 Ablation Experiments 6 Conclusion References Leveraging Tri-Planar Views and Weighted Average Fusion Technique to Classify Lung Nodule Malignancy 1 Introduction 2 Method and Workflow 2.1 Pre-trained Deep Convolutional Neural Network (DCNN) Architectures 2.2 Importance of Global Average Pooling (GAP) Layer 2.3 Proposed Weighted Average Fusion Technique 3 Dataset and Data Handling 3.1 Data Transformation and Patch Generation 3.2 Image Augmentation 4 Results and Evaluation 4.1 Evaluation Metrics 4.2 Performance Evaluation Based on Single Model 4.3 Results for Weighted Average Based Fusion Model 5 Conclusion References A Bayesian Approach to Gaussian-Impulse Noise Removal Using Hessian Norm Regularization 1 Introduction 2 Preliminaries 3 Problem Formulation 4 Proposed Methodology 4.1 Proximal Operator Associated with Function 1(T1 x) 4.2 Proximal Operator Associated with Function 3(T3 x) 5 Experimental Results 6 Conclusion References DeepTemplates: Object Segmentation Using Shape Templates 1 Introduction 2 Related Work 3 The Deep-Template Framework 3.1 Decoder (DEC) Network 3.2 Encoder (ENC) Network 3.3 End-to-End Network (E2E) 4 Implementation Details 4.1 Database Details 5 Results and Analysis 6 Conclusion and Future Work References Data-Centric Approach to SAR-Optical Image Translation 1 Introduction 2 Related Work 3 Model Components 3.1 Conditional Adversarial Network 3.2 CNN Classifier 4 Data 4.1 Sentinel Missions 4.2 Curated Dataset 5 Experiments and Results 5.1 SAR Classifier 5.2 Composite Model 5.3 Baseline Model Comparison 6 Discussion 7 Future Work References Linear and Non-Linear Filter-based Counter-Forensics Against Image Splicing Detection 1 Introduction 2 Related Works and Our Contribution 2.1 Our Contribution 3 Feature Based Image Splicing Detection 3.1 Feature Sets Explored 3.2 Adopted Model for Feature-based Splicing Detection in Images 3.3 Splicing Detection Result 4 Deep Learning Based Image Splicing Detection 4.1 Adopted Model for Deep Learning-based Image Splicing Detection 4.2 Result for Splicing Detection 5 Counter Forensic Analysis on Adopted Model 5.1 Linear Filtering 5.2 Non-Linear Filtering 6 Experiments and Results of Counter Forensics 6.1 Dataset and Implementation 6.2 Performance Evaluation Metrics 6.3 Experimental Results 7 Conclusion and Future Work References Ischemic Stroke Lesion Segmentation in CT Perfusion Images Using U-Net with Group Convolutions 1 Introduction 2 Materials and Methods 2.1 Dataset 2.2 Pre-processing and Data Augmentation 2.3 Proposed Model 2.4 Loss Function 3 Results and Discussion 3.1 Experimental Configurations 3.2 Experimental Results 4 Conclusion and Future Scope References Multi-generator MD-GAN with Reset Discriminator: A Framework to Handle Non-IID Data 1 Introduction 2 Key Related Research 3 Proposed Methodology 3.1 Training Generator Pool 3.2 Training Local Discriminator 3.3 Training Global Classifier 3.4 Analytical Validation 3.5 Evaluation Metrics 4 Experiments 4.1 Dataset 4.2 Models 4.3 Baselines for Comparison 4.4 Experiment Results 5 Conclusion and Future Scope References Video Colorization Using Modified Autoencoder Generative Adversarial Networks*-12pt 1 Introduction 2 Methodology 2.1 Generative Adversarial Networks 2.2 Autoencoders 2.3 DenseNet Architecture 3 Proposed Model 3.1 Model Workflow 4 Experimental Results and Discussion 4.1 Dataset and Implementation 4.2 Performance Measures 4.3 Results and Discussions 5 Conclusions References Real-Time Violence Detection Using Deep Neural Networks and DTW 1 Introduction 2 Related Work 3 Methodology 3.1 Overview 3.2 YoloV3 3.3 Deepsort 3.4 OpenPose 3.5 Dataset 3.6 DTW 3.7 Classifier 4 Results 5 Conclusion References Skin Disease Detection Using Saliency Maps and Segmentation Techniques 1 Introduction 2 Related Work 3 Methodology 3.1 Datasets 3.2 Data Augmentation 3.3 Preprocessing 3.4 Image Segmentation 3.5 Classification 3.6 Visualization Using Saliency Maps 4 Experimental Results 5 Conclusion References An Alternate Approach for Single Image Haze Removal Using Path Prediction*-12pt 1 Introduction 2 Proposed Work 2.1 Color Class Labeling for Path Prediction 2.2 Predicting Coloration 2.3 Alternate Method for Haze Removal 3 Results 4 Conclusion References Detecting Tropical Cyclones in INSAT-3D Satellite Images Using CNN-Based Model 1 Introduction 2 Data 3 Methodology 4 Analysis 5 Results 6 Summary and Conclusion References Two Stream RGB-LBP Based Transfer Learning Model for Face Anti-spoofing 1 Introduction 2 Related Works 3 Proposed Work 3.1 Xception Model 3.2 Local Binary Patterns (LBP) 4 Experimental Analysis 5 Conclusion References Logarithmic Progressive-SMOTE: Oversampling Minorities in Retinal Fundus Multi-disease Image Dataset 1 Introduction 1.1 Motivation 1.2 Related Work 1.3 Summary of Major Contributions 2 Proposed Method 2.1 Material 2.2 Logarithmic Progressive Synthetic Minority Oversampling Technique (LP-SMOTE) 3 Experimental Results and Discussion 4 Conclusion and Future Work References Sequence Recognition in Bharatnatyam Dance 1 Introduction 2 Related Work 2.1 Recognition of Key Postures 2.2 Recognition of Motion 2.3 Recognition of Adavus 3 Data Set 3.1 Data Set for Adavu 4 Workflow 5 Feature Extraction 5.1 Feature Extraction for Motion 5.2 Feature for Adavu 6 KP Recognition 6.1 KP Recognition: Result Discussion 7 Motion Recognition 7.1 Motion Recognition – Result Discussion 8 Adavu Recognition 8.1 Result Discussion 8.2 Performance Analysis Based on Time 9 Conclusion References Multi-modality Fusion for Siamese Network Based RGB-T Tracking (mfSiamTrack) 1 Introduction 2 Key Contributions 3 Related Work 3.1 Tracking in Thermal Imagery 3.2 Image Fusion 3.3 Semi-supervised Video Object Segmentation 4 Methodology 4.1 Siamese Network 4.2 Siamese Networks for Visual Object Tracking 4.3 Multi-modality Fusion Networks 4.4 mfSiamTrack 5 Experiments and Results 5.1 Experimental Results 6 Conclusion References Automated Detection of Changes in Built-Up Areas for Map Updating: A Case Study in Northern Italy 1 Introduction 2 State of the Art 3 The cpd4sits Routine 4 Test Sites 4.1 The Pavia, Italy Test Site 4.2 Differences with Belgian Test Site 5 Tools and Data 5.1 Ground Reference Data 5.2 Satellite Data 5.3 Parameters and Testing Combination 6 Experimental Results 7 Conclusions References Adaptive Learning for Leather Image Pre-processing Using GAN 1 Introduction 2 Review on Image Pre-processing 3 GAN-based Pre-processing 3.1 Data Used 3.2 Methodology 3.3 Implementation 4 Experimental Procedure 4.1 Subjective Validation 4.2 Objective Validation 5 Conclusion References Automated Sulcus Depth Measurement on Axial Knee MR Images 1 Introduction 2 Related Work 3 Methodology 3.1 Dataset Used 3.2 Data Augmentation 3.3 Segmentation Models Used 3.4 Model Training 3.5 Automated Sulcus Depth Measurement 4 Experimental Results 4.1 Performance Metrics for Image Segmentation 4.2 Segmentation Results 4.3 Automated Results of Sulcus Depth Measurement 5 Conclusion References LiSHT: Non-parametric Linearly Scaled Hyperbolic Tangent Activation Function for Neural Networks 1 Introduction 2 Proposed LiSHT Activation Function 3 Mathematical Analysis 4 Experimental Setup 4.1 Datasets Used 4.2 Tested Neural Networks 4.3 Training Settings 5 Results and Analysis 5.1 Experimental Results 5.2 Result Analysis 5.3 Analysis of Activation Feature Maps 5.4 Analysis of Final Weight Distribution 5.5 Analysis of Loss Landscape 6 Conclusion References Plant Disease Classification Using Hybrid Features*-12pt 1 Introduction 2 Related Works 3 Dataset 4 Methods 4.1 Handcrafted Feature Extractor (HFE) 4.2 Convolutional Feature Extractor (CFE) 4.3 Fully Convolutional Neural Network (FCNN) 4.4 Model Validation 5 Experimental Setup 6 Results 6.1 Discussion 6.2 Conclusion References Analyzing Hydro-Estimator INSAT-3D Time Series with Outlier Detection 1 Introduction 2 Related Work 3 Proposed Approach 3.1 Dataset 3.2 Big Data Techniques for Geo-Spatial Data Handling 3.3 Time Series Forecasting 3.4 SARIMA 3.5 SVR 3.6 Bi-LSTM 4 Conclusion and Future Work References Scalable Architecture for Mining Big Earth Observation Data: SAMBEO 1 Introduction 1.1 Big Earth Observation Data 2 Prevailing Architectures for Big Earth Observation Data Mining 2.1 Storage and Archival 2.2 Accessing Techniques 2.3 Processing 2.4 Analysis 2.5 Visualization 3 Challenges in Prevailing Architectures 4 Proposed Architecture 4.1 Cluster Specifications 4.2 Features of Proposed Architecture 5 Experimentation and Results 5.1 Spatial Query Performance 5.2 Temporal Query Performance 5.3 Spatio-Temporal Query Performance 6 Conclusions and Future Scope References An Efficient Deep Transfer Learning Approach for Classification of Skin Cancer Images*-12pt 1 Introduction 2 Related Work 3 Background 3.1 Deep Transfer Learning 3.2 Pre-trained Models 4 Methodology 4.1 Dataset 4.2 Preprocessing 4.3 Model Architecture and Training 4.4 Evaluation Metrics 5 Results and Discussion 6 Conclusion References Computer Vision Based Mechanism for Detecting Fire and Its Classes*-12pt 1 Introduction 2 Literature Survey 2.1 Fire and Its Classes 2.2 Fire Detection Models 3 Motivation 4 BackInTime 4.1 Caching of Frames 4.2 Classifying the Fire 5 System Implementation 5.1 Requirements 5.2 Data Collection and Preprocessing 5.3 Training 5.4 Testing 6 Results and Discussion 6.1 Fire Detection Results 6.2 Source Object Detection Results 7 Conclusion and Future Scope References A Random Forest-based No-Reference Quality Metric for UGC Videos 1 Introduction 2 Related Works 3 Proposed Methodology 3.1 Feature Extractor 3.2 Visual Quality Predictor 4 Experimental Results 4.1 Comparative Study 5 Conclusion References Endmember Extraction with Unknown Number of Sources for Hyperspectral Unmixing 1 Introduction 2 Methodology 3 Experimental Results 4 Conclusions References Advancement in Spectrum Sensing Algorithms in Cognitive Radio Architecture 1 Introduction 2 Cognitive Radio Architecture 2.1 Functionality of Cognitive Radio Networks (CRNs): 2.2 CRN Architecture 3 Spectrum Sensing in Cognitive Radio 4 FBMC Based Spectrum Sensing in Cognitive Radio 4.1 FBMC for Spectrum Sensing 5 Future Research Direction 6 Conclusion References Fast Detection and Rule Based Classification of Bharatanatyam hasta mudra 1 Introduction 2 Literature Survey 3 Preliminaries 3.1 Hand Detection Using BlazePalm Model 3.2 Hand Landmark Model 4 Proposed Methodology 4.1 Hand Detection and Keypoint Localization 4.2 Calculation of Flexion Joint Angles 4.3 Rule Based Classification 5 Experimental Results 6 Conclusion References MFNet: A Facial Recognition Pipeline for Masked Faces Using FaceNet 1 Introduction 2 Related Work 3 Dataset 4 Methodology 4.1 Data Preprocessing 4.2 Models 4.3 Training Strategy 5 QuadTriplet Loss 6 Results 7 Conclusion References Deep Learning Based Novel Cascaded Approach for Skin Lesion Analysis 1 Introduction and Related Work 2 Dataset 3 Proposed Methodology 3.1 Segmentation Approach 3.2 Classification Approach 4 Result and Discussion 5 Conclusion References Attending Local and Global Features for Image Caption Generation 1 Introduction 2 Related Work 3 The Proposed Work 3.1 Overview of Complete Architecture 3.2 Proposed Decoder Architecture 3.3 Gated Linear Unit 3.4 Multi-head Attention 4 Experiments 5 Conclusion and Future Work References A Study on an Ensemble Model for Automatic Classification of Melanoma from Dermoscopy Images 1 Introduction 1.1 Clinical Challenges 1.2 Computational Challenges 1.3 Contributions 2 Related Work 3 Methodology 3.1 Pre-processing 3.2 Classification Architecture 3.3 Test Time Augmentation 4 Experimental Results 4.1 Dataset 4.2 Hyper Parameters 4.3 Stratified K-fold Cross Validation 4.4 Evaluation Metrics 4.5 System Implementation 5 Discussions 6 Conclusions and Future Work References Challenges in Data Extraction from Graphical Labels in the Commercial Products*-12pt 1 Introduction 2 Related Works 3 Methodology 3.1 Dataset Collection and Pre-processing 3.2 Region of Interest Identification 3.3 Text Detection 4 Results and Discussions 4.1 Evaluation Metrics 4.2 Region Identification: 4.3 Text Recognition 5 Challenges in Reading Text 6 Conclusion and Future Work References A Novel Deep Learning Method for Thermal to Annotated Thermal-Optical Fused Images 1 Introduction 2 Proposed Method 2.1 Deep Network 2.2 Region of Fusion (RoF) 2.3 Database 3 Experimental Results 4 Discussion 5 Conclusion References A Compact-Structured Convolutional Neural Network for Single Image Denoising and Super-Resolution 1 Introduction 2 Related Work 3 Proposed Algorithm 3.1 Color Transformation 3.2 DnSR Architecture 3.3 Computational Complexity 3.4 Kernel Size and Parameters 4 Experimental Analysis 4.1 Dataset 4.2 Training Parameters 4.3 Comparison with the State-Of-The-Art (SOTA) Methods 4.4 Qualitative Results 4.5 Run Time 5 Conclusion References A Case Study of Rice Paddy Field Detection Using Sentinel-1 Time Series in Northern Italy*-12pt 1 Introduction 2 Data Sources 2.1 Land Cover: The DUSAF Repository 2.2 Spaceborne Radar Data 3 Data Processing 4 Conclusions References The UNICT-TEAM Vision Modules for the Mohamed Bin Zayed International Robotics Challenge 2020 1 Introduction 1.1 Challenge 1 1.2 Challenge 2 2 Related Work 2.1 Object Detection and Recognition 2.2 Object Pose Estimation 3 Method 3.1 Ball Detection 3.2 Drone Recognition 3.3 Brick Recognition 4 Experimental Settings and Results 4.1 Ball Detection 4.2 Drone Recognition 4.3 Brick Recognition 5 The Robotic Platforms 6 Conclusions References ResUNet: An Automated Deep Learning Model for Image Splicing Localization 1 Introduction 2 Related Work 3 Proposed CNN-based Spliced Region Localization Model 3.1 ResUNet-encoder 3.2 ResUNet-Decoder 4 Implementation and Results 4.1 Dataset Description 4.2 Model Training and Initialization 4.3 Evaluation Metrics 4.4 Performance Evaluation 4.5 Comparative Analysis with Existing Methods 5 Conclusion and Future Scope References Author Index