This book presents and analyzes methods to perform image co-segmentation. In this book, the authors describe efficient solutions to this problem ensuring robustness and accuracy, and provide theoretical analysis for the same. Six different methods for image co-segmentation are presented. These methods use concepts from statistical mode detection, subgraph matching, latent class graph, region growing, graph CNN, conditional encoder–decoder network, meta-learning, conditional variational encoder–decoder, and attention mechanisms. The authors have included several block diagrams and illustrative examples for the ease of readers. This book is a highly useful resource to researchers and academicians not only in the specific area of image co-segmentation but also in related areas of image processing, graph neural networks, statistical learning, and few-shot learning.
Author(s): Avik Hati, Rajbabu Velmurugan, Sayan Banerjee, Subhasis Chaudhuri
Series: Studies in Computational Intelligence, 1082
Edition: 1st ed. 2023
Publisher: Springer
Year: 2023
Language: English
Pages: 230
City: Singapore
Preface
Contents
About the Authors
Acronyms
1 Introduction
1.1 Image Co-segmentation
1.2 Image Saliency and Co-saliency
1.3 Basic Components of Co-segmentation
1.3.1 The Problem
1.4 Organization of the Monograph
1.4.1 Co-segmentation of an Image Pair
1.4.2 Robust Co-segmentation of Multiple Images
1.4.3 Co-segmentation by Superpixel Classification
1.4.4 Co-segmentation by Graph Convolutional Neural Network
1.4.5 Conditional Siamese Convolutional Network
1.4.6 Co-segmentation in Few-Shot Setting
2 Survey of Image Co-segmentation
2.1 Unsupervised Co-segmentation
2.1.1 Markov Random Field Model-Based Methods
2.1.2 Saliency-Based Methods
2.1.3 Other Co-segmentation Methods
2.2 Supervised Co-segmentation
2.2.1 Semi-supervised Methods
2.2.2 Deep Learning-Based Methods
2.3 Co-segmentation Datasets
3 Mathematical Background
3.1 Superpixel Segmentation
3.2 Label Propagation
3.2.1 Two-class Label Propagation
3.2.2 Multiclass Label Propagation
3.3 Subgraph Matching
3.4 Convolutional Neural Network
3.4.1 Nonlinear Activation Functions
3.4.2 Pooling in CNN
3.4.3 Regularization Methods
3.4.4 Loss Functions
3.4.5 Optimization Methods
3.5 Graph Convolutional Neural Network
3.6 Variational Inference
3.7 Few-shot Learning
4 Maximum Common Subgraph Matching
4.1 Introduction
4.1.1 Problem Formulation
4.2 Co-segmentation for Two Images
4.2.1 Image as Attributed Region Adjacency Graph
4.2.2 Maximum Common Subgraph Computation
4.2.3 Region Co-growing
4.2.4 Common Background Elimination
4.3 Multiscale Image Co-segmentation
4.4 Experimental Results
4.5 Extension to Co-segmentation of Multiple Images
5 Maximally Occurring Common Subgraph Matching
5.1 Introduction
5.2 Problem Formulation
5.2.1 Mathematical Definition
5.2.2 Multi-image Co-segmentation Problem
5.2.3 Overview of the Method
5.3 Superpixel Clustering
5.3.1 Feature Computation
5.3.2 Coarse-level Co-segmentation
5.3.3 Hole Filling
5.4 Common Object Detection
5.4.1 Latent Class Graph
5.4.2 Region Growing
5.5 Experimental Results
5.5.1 Quantitative and Qualitative Analysis
5.5.2 Multiple Class Co-segmentation
5.5.3 Computation Time
6 Co-segmentation Using a Classification Framework
6.1 Introduction
6.1.1 Problem Definition
6.2 Co-segmentation Algorithm
6.2.1 Mode Estimation in a Multidimensional Distribution
6.2.2 Discriminative Space for Co-segmentation
6.2.3 Spatially Constrained Label Propagation
6.3 Experimental Results
6.3.1 Quantitative and Qualitative Analyses
6.3.2 Ablation Study
6.3.3 Analysis of Discriminative Space
6.3.4 Computation Time
7 Co-segmentation Using Graph Convolutional Network
7.1 Introduction
7.2 Co-segmentation Framework
7.2.1 Global Graph Computation
7.3 Graph Convolution-Based Feature Computation
7.3.1 Graph Convolution Filters
7.3.2 Analysis of Filter Outputs
7.4 Network Architecture
7.4.1 Network Training and Testing Strategy
7.5 Experimental Results
7.5.1 Internet Dataset
7.5.2 PASCAL-VOC Dataset
8 Conditional Siamese Convolutional Network
8.1 Introduction
8.2 Co-segmentation Framework
8.2.1 Conditional Siamese Encoder-Decoder Network
8.2.2 Siamese Metric Learning Network
8.2.3 Decision Network
8.2.4 Loss Function
8.2.5 Training Strategy
8.3 Experimental Results
8.3.1 PASCAL-VOC Dataset
8.3.2 Internet Dataset
8.3.3 MSRC Dataset
8.3.4 Ablation Study
9 Few-shot Learning for Co-segmentation
9.1 Introduction
9.2 Co-segmentation Framework
9.2.1 Class Agnostic Meta-Learning
9.2.2 Directed Variational Inference Cross-Encoder
9.3 Network Architecture
9.3.1 Encoder-Decoder
9.3.2 Channel Attention Module (ChAM)
9.3.3 Spatial Attention Module (SpAM)
9.4 Experimental Results
9.4.1 PMF Implementation Details
9.4.2 Performance Analysis
9.4.3 Ablation Study
10 Conclusions
10.1 Future Work
Appendix References
