Title Page
Preface
Organization
Table of Contents
Introduction
What We Do Well
Geometric Detail for Point-like Primitives
Some Cases of Curved Primitives
Template Matching
What This Volume Describes
Shape
Shading
Grouping
Representation and Recognition
Statistics, Learning, and Recognition
An Empirical-Statistical Agenda for Recognition
What We Do Badly
Managing Information
Abstraction
Segmentation
Learning
Nagging Issues
A Brief Sketch of Bayesian Inference
How Probabilistic Models Can Address Our Problems
Using Posteriors
An Example: Colour Constancy by Sampling
The Generative Model
The Sampling Process
Experimental Examples
How Inference Methods Can Help Address Our Problems
Information Integration
Segmentation
Learning, Representation, and Primitives
Feature Selection
Summary
A Formal-Physical Agenda for Recognition
Overview
The Recognition Task
The Current State
Theory vs Empiricism
Evidence for Optimism
Summary
Shape Models and Object Recognition
Introduction
The State of the Art and Its Limitations
An Example of What Can Be Done Today
Is This Really Recognition?
What Else Then?
Generalized Cylinders
Approach
Ribbons
Generalized Cylinders
Evolving Surfaces
Background
Singularities of Evolving Surfaces
Computing the Singularities
Toward a Scale-Space Aspect Graph
Order Structure, Correspondence, and
Shape Based Categories
Introduction
Order Structure
Capturing Perceptual Shape Equivalence
Point Sets
Sets of Lines
Combination of Points and Lines
Order Structure Index from Points and Tangent Lines
Combinatorial Geometric Hashing - Computing Correspondence
Results and Discussion
Quasi-Invariant Parameterisations
and Their Applications in Computer Vision
Introduction
Semi-local Invariants
Infinitesimal Quasi-Invariance
Vector Fields of the Group
Exact Invariance
Infinitesimal Quasi-Invariance
Quasi-Invariance on Smooth Manifolds
Prolongation of Vector Fields
Quasi-Invariance on Smooth Manifolds
Quasi-Invariant Parameterisation
Affine Quasi-Invariant Parameterisation
Prolongation of Affine Vector Fields
Affine Quasi-Invariant Parameterisation
Experiments
Noise Sensitivity of Quasi Invariants
Curve Matching Experiments
Extracting Symmetry Axes
Discussion
Representations for Recognition
Under Variable Illumination
Introduction
The Illumination Cone
Illumination Cones for Convex Lambertian Surfaces
Illumination Cones for Arbitrary Objects
Illumination Cones for Non-convex Lambertian Surfaces
An Empirical Investigation: Building Illumination Cones
Dimension and Shape of the Illumination Cone
The Dimension of the Illumination Cone
The Connection between Albedo and Cone Shape
Shape of the Illumination Cone
Empirical Investigation of the Shape of the Illumination Cones
Color
Narrow-Band Cameras
Light Sources with Identical Spectra
Face Recognition Using the Illumination Cone
Constructing the Illumination Cone Representation of Faces
Recognition
Experiments and Results
Discussion of Face Recognition Results
Conclusions and Discussion
Interreflection
Effects of Change in Pose
Object Recognition
Shadows, Shading, and Projective Ambiguity
Introduction
Shadowing Ambiguity
Perspective Projection: GPBR
Orthographic Projection: GBR
Shading Ambiguity
Uniqueness of the Generalized Bas-Relief Transformation
Reconstruction from Attached Shadows
Discussion
Grouping in the Normalized Cut Framework
Introduction
Segmentation Using Normalized Cuts
The Mass-Spring Analogy
Local Image Features
Brightness, Color, and Texture
Contour Continuity
Motion and Stereo Disparity
Results
Discussion
Acknowledgements
Geometric Grouping of Repeated Elements
within Images
Introduction
The Image Relation Induced by Repetitions on a Plane
Grids
Grouping Imaged Repeated Patterns
Elation Grouping Algorithm
Grid Grouping Algorithm
Grouping Performance
Conclusions and Extensions
Constrained Symmetry for Change Detection
Introduction
Change Detection and Aerial Surveillance
Symmetry as a Generic Object Model
Recovering the Symmetry
Related Work
Unconstrained Problem
Invariance of the Ramer Approximation
Constrained Problem
Recovering the 2-d Symmetry Axis
Grouping by 2-d Symmetry
Recovering 3-d Shape from Symmetry
Further Work
Acknowledgements
Grouping Based on Coupled Di usion Maps
Introduction
Grouping as Redundancy Reduction
Computational Grouping Principles
Local Grouping
Bilocal Grouping
The Bilocal Blueprint
Bilocal Specialisations
Examples of Bilocal Grouping
Conclusions
Integrating Geometric and Photometric
Information for Image Retrieval
Introduction
Image Retrieval Based on Intensity Invariants
Interest Points
Intensity Invariants
Retrieval Algorithm
Semi-local Constraints
Experimental Results
Curve Matching
Basic Curve Matching Algorithm
Wide Base Line Matching Algorithm
Image Matching Using Curve Verification
Discussion and Extensions
Towards the Integration of Geometric and
Appearance-Based Object Recognition
Overview
Edge-Based Regionization
Regions Derived from Edgel Boundaries
Intensity Models
Variation of Intensity Models with Viewpoint
Remarks
Affine Indexing
Initial Thoughts about Intensity Indexing
Discussion
Recognizing Objects Using Color-Annotated
Adjacency Graphs
Outline
Extracting Object Faces from Images
Detecting Approximate Region Boundaries
Estimating Initial Uniform Regions: Constrained Triangulation
Extracting Object Faces: Region Merging
Deriving Graph Representations of Objects
The Three-Tier Matching Method
Local Matching
Neighborhood Matching
Graph Matching
Solution Criteria
Binary Quadratic Formulation
Approximating the Clique with the Largest Degree of Mutual Compatibility
Results
Conclusion
A Cooperating Strategy for Objects Recognition
Introduction
The Early Vision Phase
Object Detection
Snakes and Segmentation
Object Axial Symmetries and Features Extraction
The Object Recognizer
Experiments and Discussion
Model Selection for Two View Geometry:
A Review
Introduction
Putative Motion Models
Maximum Likelihood Estimation
Model Selection---Hypothesis Testing
{sc AIC} for Model Selection
Bayes Factors
Calculating Bayes Factors
Modified {sc BIC} for Least Squares Problems
The Quest for the Universal Prior: {sc MDL}
Bayesian Model Selection and Model Averaging
Results
Discussion
Conclusion
Finding Objects by Grouping Primitives
Background
Primitives, Segmentation, and Implicit Representations
Body Plans - Interim Results on Implicit Representations
Learning Assembly Processes from Data
Results
Shading Primitives, Shape Representations, and Clothing
Grouping Folds Using a Simple Buckling Model
Grouping Folds by Sampling
Choosing Primitives and Building Representations
Conclusions
Object Recognition with Gradient-Based
Learning
Learning the Right Features
Shape Recognition with Convolutional Neural Networks
Convolutional Networks
LeNet-5
An Example: Recognizing Handwritten Digits
Results and Comparison with Other Classifiers
Invariance and Noise Resistance
Multiple Object Recognition with Space Displacement Neural Networks
Interpreting the Output of an SDNN
Experiments with SDNN
Face Detection and Spotting with SDNN
Graph Transformer Networks
Word Recognition with a Graph Transformer Network
Gradient-Based Training of a GTN
Conclusion
Author Idex
