Computer Vision -- ECCV 2010: 11th European Conference on Computer Vision, Heraklion, Crete, Greece, September 5-11, 2010, Proceedings, Part IV (Lecture Notes in Computer Science, 6314)

The 2010 edition of the European Conference on Computer Vision was held in Heraklion, Crete. The call for papers attracted an absolute record of 1,174 submissions. We describe here the selection of the accepted papers: Thirty-eight area chairs were selected coming from Europe (18), USA and Canada (16), and Asia (4). Their selection was based on the following criteria: (1) Researchers who had served at least two times as Area Chairs within the past two years at major vision conferences were excluded; (2) Researchers who served as Area Chairs at the 2010 Computer Vision and Pattern Recognition were also excluded (exception: ECCV 2012 Program Chairs); (3) Minimization of overlap introduced by Area Chairs being former student and advisors; (4) 20% of the Area Chairs had never served before in a major conference; (5) The Area Chair selection process made all possible efforts to achieve a reasonable geographic distribution between countries, thematic areas and trends in computer vision. EachArea Chair was assigned by the Program Chairs between 28–32 papers. Based on paper content, the Area Chair recommended up to seven potential reviewers per paper. Such assignment was made using all reviewers in the database including the conflicting ones. The Program Chairs manually entered the missing conflict domains of approximately 300 reviewers. Based on the recommendation of the Area Chairs, three reviewers were selected per paper (with at least one being of the top three suggestions), with 99. Title Page Preface Organization Table of Contents – Part IV Kernel Sparse Representation for Image Classification and Face Recognition Introduction Kernel Sparse Representation and Implementation Kernel Sparse Representation Implementation Application I: Kernel Sparse Representation for Image Classification Sparse Coding for Codebook Generation Maximum Feature Pooling and Spatial Pyramid Matching Based Image Representation KSRSPM – An Generalization of Efficient Matching Kernel Experiments Application II: Kernel Sparse Representation for Face Recognition Sparse Coding for Face Recognition Kernel Sparse Representation for Face Recognition Evaluation on Extended Yale B Database Conclusion References Every Picture Tells a Story: Generating Sentences from Images Introduction Approach Mapping Image to Meaning Image Potentials Sentence Potentials Learning Evaluation Dataset Inference Matching Out of Vocabulary Extension Experimental Settings Mapping to the Meaning Space Results Mapping Images to Meanings Annotation: Generating Sentences from Images Illustration: Finding Images Best Described by Sentences Out of Vocabulary Extension Discussion and Future Work References An Eye Fixation Database for Saliency Detection in Images Introduction Eye Fixation Database Data Collection Protocol Image Content Analysis of Visual Attention Characteristics Enhancing Active Image Segmentation with Multiple Fixations Algorithm Analysis Results and Discussion Conclusion and Future Work References Face Image Relighting Using Locally Constrained Global Optimization Introduction Related Work A Relighting Approach and Its Problems The Proposed Method Experimental Results and Comparisons Conclusions References Correlation-Based Intrinsic Image Extraction from a Single Image Introduction Basic Cues Steerable Filter Based Feature Extraction Extraction of AM Extraction of TM Extraction of HM Reconstruction of Shading and Reflectance Post-processing of the Reconstructed Images DC Component of Shading Image Compensation of Reflectance in Regions of Deep Shadow Experimental Results Test Set Evaluation of Extraction Method and Cue Combinations Enhancement of Reconstructed Images Evaluation on Natural Images Conclusion References ADICT: Accurate Direct and Inverse Color Transformation Introduction Previous Work Algorithm Color Transfer Function Color Transformation Implementation Results Discussion Conclusion References Real-Time Specular Highlight Removal Using Bilateral Filtering Introduction Algorithm Reflection Model Highlight Removal Using Bilateral Filter Experimental Results Conclusions References Learning Artistic Lighting Template from Portrait Photographs Introduction Feature Design Template Learning Inference for Classification and Assessment Classification Numerical Assessment Experiments Data Collection The Learned Template Classification Results Numerical Assessment Results Discussions Conclusion References Photometric Stereo from Maximum Feasible Lambertian Reflections Introduction Max FS Formulation for Photometric Stereo Lambertian Photometric Stereo Maximum Feasible Subsystem Problem Max FS ( Big-M MILP ) Formulation for Photometric Stereo Experimental Results Experiment with a Synthetic Sphere Experiments with Real Objects Conclusion References Part-Based Feature Synthesis for Human Detection Introduction Overview and Related Work Part Based Feature Generation Predictive Feature Selection Analysis Empirical Evaluation PFS Evaluation Feature Synthesis for Human Detection Conclusions and Future Work References Improving the Fisher Kernel for Large-Scale Image Classification Introduction The Fisher Vector Improving the Fisher Vector L2 Normalization Power Normalization Spatial Pyramids Evaluation of the Proposed Improvements Experimental Setup PASCAL VOC 2007 CalTech 256 Large-Scale Experiments: ImageNet and Flickr Groups Conclusion References Max-Margin Dictionary Learning for Multiclass Image Categorization Introduction Related Work Max-Margin Dictionary Learning Problem Formulation MMDL Base Classifier and Aggregation Strategy Time Complexity Experiments Object Localization Scene Category Classification Conclusion References Towards Optimal Naive Bayes Nearest Neighbor Introduction Parametric NBNN Classification Initial Formulation of NBNN Affine Correction of Nearest Neighbor Distance for NBNN Multi-channel Image Classification Parameter Estimation Multi-channel Classification by Detection Experiments Single-Channel Classification Radiometry Invariance Multi-channel Classification Classification by Detection Conclusion References Weakly Supervised Classification of Objects in Images Using Soft Random Forests Introduction Decision Trees and Random Forest Supervised Decision Trees and Random Forests Weakly Supervised Learning of Soft Decision Trees Iterative Classification Naive Iterative Procedure Randomization-Based Iterative Procedure without over Training Application to Semi-supervised Learning Related Work Self Training with Soft Random Forests Experiments Simulation Protocol Experiments on Weakly Supervised Dataset Application to Fish School Classification in Sonar Images Semi-supervised Experiments Conclusion References Learning What and How of Contextual Models for Scene Labeling Introduction Related Work Overview Mathematical Formulation Iterative Approach Inference Experimental Results References Adapting Visual Category Models to New Domains Introduction Related Work Domain Adaptation Using Regularized Cross-Domain Transforms Domain Adaptation Using Metric Learning A Database for Studying Effects of Domain Shift in Object Recognition Experiments Conclusion References Improved Human Parsing with a Full Relational Model Introduction Related Work Method Searching a Full Energy Model Training a Full Energy Model Part Detectors Features Experimental Results Dataset Results Discussion References Multiresolution Models for Object Detection Introduction Related Work Multiresolution Models Fixed-Resolution Models Multiple Fixed-Resolution Models Multiscale Multiresolution Models Multiresolution Part Models Latent Multiresolution Part Models Multiresolution Contextual Models Experimental Results Benchmark Results Diagnostic Experiments Conclusion References Accurate Image Localization Based on Google Maps Street View Introduction Google Maps Street View Dataset Single Image Localization Confidence of Localization Image Group Localization Experiments Single Image Localization Results Image Group Localization Results Conclusion References A Minimal Case Solution to the Calibrated Relative Pose Problem for the Case of Two Known Orientation Angles Introduction Related Work Estimating the Essential Matrix for the Case of Two Known Orientation Angles The Linear 5-Point Algorithm The 4-Point Algorithm The 3-Point Minimal Case Algorithm Degeneracies Experiments Synthetic Data Real Data from N900 Smartphone Relative Pose Estimation for MAV Camera-IMU System Conclusion References Bilinear Factorization via Augmented Lagrange Multipliers Introduction Related Work Problem Statement The BALM Algorithm Solving for (6) Solving for (7) Solving for (8) Example 1: BALM for Non-rigid SfM NRSfM Manifold Projector Example 2: BALM for Photometric Stereo Photometric Stereo Manifold Projector Experiments Synthetic Experiments: NRSfM Real Data: NRSfM Real Data: Photometric Stereo Conclusions References Piecewise Quadratic Reconstruction of Non-Rigid Surfaces from Monocular Sequences Introduction Related Work Piecewise Non-Rigid Structure from Motion Division of the Surface into Patches Reconstruction of Individual Patches Quadratic Deformation Model Non-linear Optimization Estimation of the Rest Shape From Local Patches to a Global Reconstruction Connecting Patches Final Optimization Experiments Conclusions and Future Work References Extrinsic Camera Calibration Using Multiple Reflections Introduction Related Work Problem Formulation Measurement Model Camera-to-Base Transformation Analytical Solution Relationship to PnP Analytical Solution for the Camera and Mirror Configurations Analytical Solution for Scene Reconstruction MLE Refinement of Analytical Solutions Simulations Experiments Conclusions and Future Work References Probabilistic Deformable Surface Tracking from Multiple Videos Introduction Related Works Method Parametrization and Deformation Framework Problem Formulation Bayesian Model Expectation-Maximization Results Multi-object Tracking and Outlier Rejection Human Performance Capture Discussion Conclusion References Theory of Optimal View Interpolation with Depth Inaccuracy Introduction Background Theory View Interpolation with Disparity Error MSE of View Interpolation Derivation of Optimal View Interpolation Examples Relation with Linear Interpolation Experiment Conclusions References Practical Methods for Convex Multi-view Reconstruction Introduction Related Work Global Optimization in Multiple View Geometry Non-smooth Convex Optimization and Proximal Methods Convex $L_1$ Reconstruction with Known Rotations Our Approach The Cost Functions Application to Multi-view Reconstruction Numerical Scheme Numerical Results Conclusion References Building Rome on a Cloudless Day Introduction Previous Work The Approach Appearance-Based Clustering with Small Codes Geometric Verification Local Iconic Scene Graph Reconstruction Dense Geometry Estimation Conclusions References Camera Pose Estimation Using Images of Planar Mirror Reflections Introduction Projection Model for Images of Planar Mirror Reflections Projection of a 3D Point Symmetry Matrices The Virtual Camera Assumptions and Problem Formulation Searching for the Mirror Planes Geometric Properties Linear Constraints Determining the Real Camera from $N$ ≥ 3 Virtual Cameras The System of Linear Equations Outline of the Algorithm Singular Configurations Performance Evaluation with Synthetic Data Experiments in Extrinsic Calibration Conclusions References Element-Wise Factorization for N-View Projective Reconstruction Introduction Element-Wise Factorization Preliminaries Element-Wise Factorization Implementation Rank Minimization Trace Minimization Extensions Dealing with Missing Data Dealing with Pure Outliers Experimental Results Synthetic Experiments Real Image Experiments Conclusion References Learning Relations among Movie Characters: A Social Network Perspective Introduction Social Network Representation Learning Social Networks Scene Level Features and Scene Characterization Learning Inter-character Affinity Social Network Analysis Experiments Conclusions and Future Work References What, Where and How Many? Combining Object Detectors and CRFs Introduction Related Work CRFs and Detectors CRFs for Labelling Problems Detectors in CRF Framework Inference for Detector Potentials Experimental Evaluation CRF Framework Detection-Based Potentials Results Summary References Visual Recognition with Humans in the Loop Introduction Related Work Visual Recognition with Humans in the Loop Incorporating Computer Vision Modeling User Responses Datasets and Implementation Details Birds-200 Dataset Animals with Attributes Implementation Details and Parameter Settings Experiments Measuring Performance Results Conclusion References Localizing Objects While Learning Their Appearance Introduction The CRF Model to Localize a New Class Localization and Learning Localization Learning Adaptation Class-Specific Appearance Models Υf Class-Specific Shape Model Π Image Responsibilities ρn Unary Appearance Cue Weights αΥf Pairwise Appearance Cue Weights αΓf Generic Knowledge: Initializing Θ Objectness Ω Pairwise Shape Similarity Λ Pairwise Appearance Similarity Γf Weights α Kernel of the SVMs Υf Percentage κ of Images Appearance Cues Experiments: WS Localization and Learning Localizing Objects in Their Weakly Supervised Training Images Experiments: Localizing Objects in New Test Images Conclusion References Monocular 3D Scene Modeling and Inference: Understanding Multi-Object Traffic Scenes Introduction Related Work Single-Frame 3D Scene Model Inference Framework Proposal Moves Projective 3D to 2D Marginalization Multi-frame Scene Model and Inference Multi-frame 3D Scene Tracklet Model Long Term Data Association with Scene Tracking Datasets and Implementation Details Experimental Results Conclusion References Blocks World Revisited: Image Understanding Using Qualitative Geometry and Mechanics Introduction Overview Block Representation Representing Relationships Geometric and Mechanical Constraints Assembling Blocks World: Interpretation by Synthesis Initialization Searching Block Configurations Evaluating Proposals Estimating Geometry Estimating Physical Stability Extracting Depth Constraints Creating Split and Merge Proposals Experimental Results References Discriminative Learning with Latent Variables for Cluttered Indoor Scene Understanding Introduction Related Work Model Learning and Inference Learning Approximate Inference Priors and Features Experimental Results Conclusion References Visual Tracking Using a Pixelwise Spatiotemporal Oriented Energy Representation Introduction Technical Approach Features: Spatiotemporal Oriented Energies Target Representation Robust Motion Estimation Template Adaptation Empirical Evaluation Discussion and Summary References A Globally Optimal Approach for 3D Elastic Motion Estimation from Stereo Sequences Introduction Related Work Proposed Approach and Contribution Symmetric Hidden Markov Models and Problem Formulation Basic Idea Symmetric Hidden Markov Models Problem Formulation Inference and Optimization Inference Under EM Algorithm Outlier and Missing Data Handling Experimental Results and Discussion Implementation and Datasets Results and Evaluation Conclusions References Occlusion Boundary Detection Using Pseudo-depth Introduction Background Step 1: Motion Flow Estimation Step 2: Dense Pseudo-depth Estimation Pseudo-depth from Motion Weak Smoothness to Improve the Pseudo-Depth Estimates Inferring Pseudo-depth Using the Weak Smoothness MRF Step 3: Occlusion Boundary Detection Experiments Dataset and Setup Experimental Results Conclusion and Discussion References Multiple Target Tracking in World Coordinate with Single, Minimally Calibrated Camera Introduction Related Work Multi-target Tracking Model Overall Method Track Initiation, Termination and Correspondence Camera Model and KLT Features Target Class Model Sequential Tracking Model with Independent Assumption From Independent to Joint Target Model Tracking Multi-target by MCMC Particle Filter Experimental Results and Implementation Details Conclusion References Joint Estimation of Motion, Structure and Geometry from Stereo Sequences Introduction A Scene Flow Model for Uncalibrated Stereo Sequences Data Constraints Occlusion Scores Epipolar Constraints Smoothness Constraints Linearisation and Normalisation Linearisation in the Data Term Treatment of the Epipolar Term Constraint Normalisation Minimisation and Numerical Solution Experiments Conclusions References Dense, Robust, and Accurate Motion Field Estimation from Stereo Image Sequences in Real-Time Introduction Related Work Outline Two-Frame Motion Field Estimation Combination of Optical Flow and Stereo Variational Scene Flow Temporal Integration of the Motion Field Model Filtered Tracks and Stereo: 6D-Vision Filtered Dense Optical Flow and Stereo: Dense6D Filtered Variational Scene Flow Evaluation Evaluation with Ground Truth Information Real World Results Conclusions References Estimation of 3D Object Structure, Motion and Rotation Based on 4D Affine Optical Flow Using a Multi-camera Array Introduction Approach Contribution ModelDerivation Surface Patch Model Rotation Projective Camera Model Pixel-Centered View Projecting the Pixel Grid to the Surface Brightness Change Model A 4D-Affine Model The Range Constraint, Zt, b4, and Why (8) Still Holds under Rotation Parameter Estimation Experiments Sinusoidal Pattern Synthetic Cube Plant Leaf Summary and Conclusions References Efficiently Scaling Up Video Annotation with Crowdsourced Marketplaces Introduction Mechanical Turk User Interface Dense Labeling Protocol User Instructions Video Server Cloud Tracking and Interpolation Linear Interpolation Discriminative Object Templates Constrained Tracking Efficient Optimization Results Diminishing Returns CPU vs. Human Cost Performance Cost Trade-Off Conclusion References Robust and Fast Collaborative Tracking with Two Stage Sparse Optimization Introduction Related Work Tracking with Two Stage Sparsity Bayesian Tracking Framework Two Stage Sparse Representation Experiments Visual Evaluation of Comparative Experiment Results Quantitative Evaluation of Comparative Experimental Results Conclusion References Nonlocal Multiscale Hierarchical Decomposition on Graphs Introduction Multiscale Hierarchical Decomposition of Images Multiscale Hierarchical Decomposition on Graphs Digital Variational Framework Numerical Resolution Digital Multiscale Hierarchical Decomposition on Graphs Parameter Selection Results Images Meshes and Point Clouds Conclusion References Adaptive Regularization for Image Segmentation Using Local Image Curvature Cues Introduction Methods Local Image Curvature Cue Curvature-Based Regularization Incorporation of Texture Cue Structural Cue Modulated Graph Cuts Segmentation Structural Cue Modulated Active Contours Segmentation Results and Discussion Conclusion References A Static SMC Sampler on Shapes for the Automated Segmentation of Aortic Calcifications Introduction Problem Definition Static SMC Sampler Approach Static SMC Sampler on Shapes Artificial Distributions Forward Markov Kernel Backward Markov Kernel Resampling Evaluation Vertebrae Segmentation Aorta Segmentation Discussion Conclusion References Fast Dynamic Texture Detection Introduction Proposed Methods Optical-Flow-Based DT Detection (OFDT) Motion Outliers-Based DT Detection (MODT) Comparison of OFDT and MODT The Approach by Fazekas et al. [25] Results Synthetic Sequences Real Sequences Conclusions References Finding Semantic Structures in Image Hierarchies Using Laplacian Graph Energy Introduction Laplacian Graph Energy as a Complexity Measure Using Laplacian Graph Energy to Filter Hierarchies Results Application to Tracking Conclusion References Semantic Segmentation of Urban Scenes Using Dense Depth Maps Introduction Related Work Overview and Contributions DepthMapsRecovery Semantic Segmentation from Dense Depth Maps Image Over-Segmentation Features from Dense Depth Map Randomized Decision Forest Graph-Cut Based Optimization Temporal Multi-view Fusion Cross Training and Testing Experiments Evaluation Using the CamVid Database Cross Training Test Conclusion References Tensor Sparse Coding for Region Covariances Introduction Region Covariance Descriptors Problem Statement Approach The LogDet Divergence Formulation The MAXDET Problem Experiments Numerical Example Classification Experiments Conclusions and Future Work References Improving Local Descriptors by Embedding Global and Local Spatial Information Introduction Outline of the Proposed Method Local Spatial Information Global Spatial Information Classifier Experiment Setup Scene Classification Object Recognition Conclusions References Sharon Alpert, Meirav Galun, Boaz Nadler, and Ronen Basri Spatial Statistics of Visual Keypoints for Texture Recognition Introduction Overview of Proposed Method Co-occurrence Statistics of Visual Keypoints Spatial Point Process Descriptive Statistics of the Spatial Patterns of Visual Keypoints Scaling Effects Dimensionality of the Feature Space Texture Classification Experiments UIUC Texture Classification Real Sonar Textures Discussion and Future Developments References BRIEF: Binary Robust Independent Elementary Features Introduction Related Work Method Smoothing Kernels Spatial Arrangement of the Binary Tests Distance Distributions Results Conclusion References Multi-label Feature Transform for Image Classifications Introduction Multi-label Feature Transform Outlines of MLFT Approach Implicit Construction of Label-Augmented Feature Vector Multi-label Kernel Laplacian Embedding PCA Laplacian Embedding Kernel Laplacian Linear Embedding (KLE) Label Correlation Enhanced Kernel Laplacian Embedding Correlative Kernel Transformation Initialization of Unlabeled Data Motivation and Formulation of Label Correlations Experimental Evaluations Evaluation Metrics for Multi-label Classification Multi-label Classification Performance Label Enhancement in Multi-label Classification Conclusions References Author Index (5) The Area Chair selection process made all possible efforts to achieve a reasonable geographic distribution between countries, thematic areas and trends in computer vision. Each Area Chair was assigned by the Program Chairs between 28–32 papers.