Computer Vision: Detection, Recognition and Reconstruction (Studies in Computational Intelligence (285), Band 285)

Description

Computer vision is the science and technology of making machines that see. It is concerned with the theory, design and implementation of algorithms that can automatically process visual data to recognize objects, track and recover their shape and spatial layout. The International Computer Vision Summer School - ICVSS was established in 2007 to provide both an objective and clear overview and an in-depth analysis of the state-of-the-art research in Computer Vision. The courses are delivered by world renowned experts in the field, from both academia and industry, and cover both theoretical and practical aspects of real Computer Vision problems. The school is organized every year by University of Cambridge (Computer Vision and Robotics Group) and University of Catania (Image Processing Lab). Different topics are covered each year. A summary of the past Computer Vision Summer Schools can be found at: http://www.dmi.unict.it/icvss This edited volume contains a selection of articles covering some of the talks and tutorials held during the first two editions of the school on topics such as Recognition, Registration and Reconstruction. The chapters provide an in-depth overview of these challenging areas with key references to the existing literature. Title Page Editors’ Biographies Preface Contents List of Contributors Is Human Vision Any Good? Introduction Frameworks The Spatial Framework The Photometric Framework A Case Study: ``Shape from Shading'' The So Called ``Shape from Shading Problem'' Setting up the Problem The Local Shape from Shading Problem Final Remarks References Knowing a Good Feature When You See It: Ground Truth and Methodology to Evaluate Local Features for Recognition Introduction Empirical Studies of Local Features Some Nomenclature Constructing a Rigorous Ground Truth Modeling the Detector Modeling Correspondences Ground-Truth Correspondences Comparing Ground-Truth and Real-World Correspondences The Data Learning to Compare Invariant Features Wide Baseline Motion Statistics Learning to Rank Matches Learning to Accept Matches Discussion Appendix References Dynamic Graph Cuts and Their Applications in Computer Vision Introduction Markov and Conditional Random Fields Graph Cuts for Energy Minimization The st-Mincut Problem Formulating the Max-Flow Problem Augmenting Paths, Residual Graphs Minimizing Functions Using Graph Cuts Minimizing Dynamic Energy Functions Using Dynamic Graph Cuts Dynamic Computation Energy and Graph Reparameterization Recycling Computation Updating Residual Graphs Computational Complexity of Update Operations Improving Performance by Recycling Search Trees Reusing Search Trees Tree Recycling for Dynamic Graph Cuts Dynamic Image Segmentation CRFs for Image Segmentation Image Segmentation in Videos Experimental Results Reusing Flow vs. Reusing Search Trees Simultaneous Segmentation and Pose Estimation of Humans Pose Specific CRF for Image Segmentation Formulating the Pose Inference Problem Experiments Shape Priors for Reconstruction Discussion Summary and Future Work Measuring Uncertainty in Graph Cut Solutions Preliminaries Computing Min-Marginals Using Graph Cuts Min-Marginals and Flow Potentials Extension to Multiple Labels Minimizing Energy Function Projections Using Dynamic Graph Cuts Computational Complexity and Experimental Evaluation Applications of Min-Marginals References Discriminative Graphical Models for Context-Based Classification Contextual Dependencies in Images The Nature of Contextual Interactions Markov Random Field (MRF) Conditional Random Field (CRF) Association Potential Interaction Potential Parameter Learning and Inference Maximum Likelihood Parameter Learning Inference Extensions Multiclass CRF Hierarchical CRF Applications Man-Made Structure Detection Image Classification and Contextual Object Detection RelatedWork and Further Readings References From the Subspace Methods to the Mutual Subspace Method Introduction The Subspace Methods A Brief History of the Subspace Methods Basic Idea Subspace Construction The Mutual Subspace Method Basic Idea Application to Chinese Character Recognition Application to Face Recognition Application to 3-D Face Recognition Conclusion References What, Where and Who? Telling the Story of anImage by Activity Classification, Scene Recognition and Object Categorization Introduction and Motivation Overall Approach Literature Review The Integrative Model Labeling an Unknown Image Learning the Model System Implementation Experiments and Results Dataset Experimental Setup Results Conclusion References Semantic Texton Forests Introduction RelatedWork Randomized Decision Forests Training the Forest Experiments Image Categorization Tree Histograms and Pyramid Matching Categorization Results Semantic Segmentation Soft Classification of Pixels Image-Level Semantic Constraints Categorization Results The Image Level Prior Compositional Constraints Experiments MSRC21 Dataset VOC 2007 Segmentation Dataset Discussion References Multi-view Object Categorization and Pose Estimation Introduction Literature Review The Part-Based Multi-view Model Overview Canonical Parts and Linkage Structure Building the Model Extract Features Form Parts Find Canonical Parts Candidates Create the Model View Synthesis Representing an Unseen View Recognizing Object Class in Unseen Views Extract Features and Get Part Candidates Recognition Procedure: First Step Recognition Procedure: Second Step Experiments and Results Experiment I: Comparison with Thomas et al. [63] Experiment II: Detection and Pose Estimation Results on the Dataset in [54] Experiment III: Detection and Pose Estimation Results on the Dataset in [55] Conclusion References A Vision-Based Remote Control Introduction PriorWork Hand Tracking for Human Computer Interfaces Commercial Gesture Interface Systems Visual Tracking of a Single Object Tracking with Multiple Observers Observation Models Evaluating Single Observers Evaluating Multiple Observers Parallel Evaluation Cascaded Evaluation Dynamic Model Discussion Experimental Results Individual Observers Observer Combinations Tracker Evaluation on Selected Combinations Gesture Interface System Visual Attention Mechanism Tracking Mechanism Selection Mechanisms Summary and Conclusion References Multi-view Multi-object Detection and Tracking Introduction Problem Formulation Calibration and Fusion Single-Level Homography Multi-Level Homography Track-First Approaches Independent Tracking Collaborative Tracking Fuse-First Approaches Detection-Based Tracking Track-Before-Detect Conclusions References Shape from Photographs: A Multi-view Stereo Pipeline Introduction Multi-view Stereo Pipeline: From Photographs to 3D Models Computing Photo-Consistency from a Set of Calibrated Photographs Normalized Cross Correlation for Depth-Map Computation Depth Map Estimation Photo-Consistency 3D Map from a Set of Depth-Maps Extracting a 3D Surface from a 3D Map of Photo-Consistency Multi-view Stereo Using Multi-resolution Graph-Cuts Discontinuity Cost from a Set of Depth-Maps Graph Structure Labeling Cost from a Set of Depth-Maps Probabilistic Fusion of Depth Sensors Deformable Models Experiments Depth Map Evaluation Multi-view Stereo Evaluation Digitizing Works of Art Discussion Appendix References Practical 3D Reconstruction Based on Photometric Stereo Introduction Photometric Stereo with Coloured Light Classic Three-Source Photometric Stereo Multi-spectral Sources and Sensors Calibration Comparison with Photometric Stereo The Problem of Shadows Facial Capture Experiments Related Work Multi-view Photometric Stereo Related Work Algorithm Experiments References Index