Computer Vision -- ECCV 2010: 11th European Conference on Computer Vision, Heraklion, Crete, Greece, September 5-11, 2010, Proceedings, Part V (Lecture Notes in Computer Science, 6315)

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. Each Area 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 V Spotlights and Posters W2 Towards Computational Models of the Visual Aesthetic Appeal of Consumer Videos Introduction Previous Work Ground Truth Data Collection Feature Computation Frame-Level Features Microshot and Video-Level Features Experimental Results Discussion Conclusions and Future Work References Object Recognition Using Junctions Introduction Junction Features Junction Descriptors Junction Descriptors for Edge Maps Detection Junction Classification Graph Model Experiments Conclusions and Future Work References Using Partial Edge Contour Matches for Efficient Object Category Localization Introduction Related Work Partial Shape Matching for Object Detection Fragment Description Fragment Matching and Merging Hypothesis Voting Experiments ETHZ Shape Classes INRIA Horses Conclusion References Active Mask Hierarchies for Object Detection Introduction Related Work Active Mask Hierarchies Active Mask Hierarchies and Latent Structural SVM The Representation: Hierarchical Model and Feature Kernels Optimization by CCCP Detection: Dynamic Programming Experiments The Detection Results on the PASCAL Dataset Active Mask Hierarchies, Spatial Pyramid and Part-Based Model Benefit of Shape Masks Weights of HOGs and HOWs Implementation Details Conclusion References From a Set of Shapes to Object Discovery Introduction Image Representation Using Shapes and Shape Description Constructing the Graph of Pairs of Image Contours Coordinate-Ascent Swendsen-Wang Cut Results Conclusion References What Does Classifying More Than 10,000 Image Categories Tell Us? Introduction Related Work Datasets Procedure Computation Matters Size Matters Density Matters Hierarchy Matters Conclusion References Modeling and Analysis of Dynamic Behaviors of Web Image Collections Introduction Related Work Network Construction by Sequential Monte Carlo Image Description and Similarity Measure Problem Statement Network Construction Using Sequential Monte Carlo Analysis and Results Flickr Dataset Evolution of Subtopics Comparison with Text Analysis Temporal Association for Classification Discussion References Non-local Characterization of Scenery Images: Statistics, 3D Reasoning, and a Generative Model Introduction Related Work Observations and Evidence The General Shape of Background vs. Foreground Objects The Top-Down Order of Background Objects Contours Separating Background Regions Why Are Background Regions Horizontal? A 3D Analysis Flatland Land Cover The World Is Wrinkled: Ground Elevation and Slope Statistics A Generative Model A Markov Network for Modeling the Top-Down Label Order A Normal Distribution for the Height Covered by Each Region Modeling Background Contours with PCA Generative Model Demonstration Discussion References Efficient Highly Over-Complete Sparse Coding Using a Mixture Model Introduction Related Works and Motivations Sparse Coding for Image Classification Local Coordinate Coding Motivation Sparse Coding Using a Mixture Model The Model Learning Algorithm Practical Implementation Image Encoding Experimental Validation PASCAL Datasets Implementation Details Classification Results Conclusion and Future Work References Attribute-Based Transfer Learning for Object Categorization with Zero/One Training Example Introduction Related Work Algorithms Background Attribute Model and Target Classifier Knowledge Transfer by Synthesis of Training Examples Knowledge Transfer by Informative Parameter Priors Experiments Data Set and Image Features Experiment Setup and Implementation Details Results Conclusion and Future Work References Image Classification Using Super-Vector Coding of Local Image Descriptors Introduction Overview of Our Approach TheMethod Descriptor Coding Spatial Pooling Image Classification Discussion and Further Improvement Experiments Comparison of Nonlinear Coding Methods Comparison with State-of-the-Art Results Impact of Codebook Size Visualization of the Learned Patch-Level Function Conclusion References A Discriminative Latent Model of Object Classes and Attributes Introduction ModelFormulation Learning Objective Non-convex Cutting Plane Training Attribute Relation Graph Other Loss Functions Experiments Conclusion References Seeing People in Social Context: Recognizing People and Social Relationships Introduction Related Work Approach Learning the Model with EM Inference Implementation Details Experiments Recognizing People with Social Relationships Recognizing Social Relationships in Novel Image Collections Conclusions References Discovering Multipart Appearance Models from Captioned Images Introduction Related Work Images, Parts and Multipart Models Discovering Parts Model Initialization through Image Pair Sampling Part Coverage Objective Building Multipart Models Detecting Duplicate Parts Locating Part Detections Choosing Initial Multipart Models Refinement and Expansion of Multipart Models Detecting Multipart Models Results Experiments on the HOCKEY Data Set Experiments on the LANDMARK Data Set Conclusions References Voting by Grouping Dependent Parts Introduction Voting Methods and Object Detection Hough Voting with Independent Parts Key Points of Our Method Grouping, Voting, and Correspondences Joint Objective Function for Grouping, Voting, and Correspondences Joint Optimization of Groups, Votes, and Correspondences Hough Voting with Groups Bottom-Up Grouping Hypothesis Verification Experiments ETHZ Shape Dataset – Performance Analysis INRIA Horse Dataset – Performance Analysis Discussion References Superpixels and Supervoxels in an Energy Optimization Framework Introduction Superpixel Segmentation Energy Minimization with Graph Cuts Compact Superpixels Variable Patch Superpixels Constant Intensity Superpixels Supervoxel Segmentation Experimental Results Application to Salient Object Segmentation Future Work References Segmentation Convex Relaxation for Multilabel Problems with Product Label Spaces Introduction The Multi-labeling Problem Discrete Approaches Continuous Approaches Contribution: Product Label Spaces Relaxations for Product Label Spaces Product Label Spaces Convex Relaxation Numerical Method Obtaining a Solution to the Original Problem Regularization Experiments Multi-label Segmentation Depth and Occlusion Map Optic Flow Conclusion References Graph Cut Based Inference with Co-occurrence Statistics Introduction CRFs and Co-occurrence Incorporating Co-occurrence Potentials PriorWork Inference on Global Co-occurrence Potentials αβ-Swap Moves α-Expansion Moves Experiments Conclusion References Ambrosio-Tortorelli Segmentation of Stochastic Images Introduction Stochastic Images Polynomial Chaos Expansion Polynomial Chaos for Stochastic Images From Samples to Input Distributions A Phase Field Model for Segmentation on Stochastic Images Classical Mumford-Shah and Ambrosio-Tortorelli Segmentation Ambrosio-Tortorelli Segmentation on Stochastic Images Weak Formulation and Discretization Generalized Spectral Decomposition Results Street Image Data Set Ultrasound Samples Conclusions References Multiple Hypothesis Video Segmentation from Superpixel Flows Introduction AnOverviewofMHVS Enumeration and Scoring of Hypotheses Hypotheses Competition Experimental Results Discussion and Future Work References Object Segmentation by Long Term Analysis of Point Trajectories Introduction Related Work Point Tracking and Affinities between Trajectories Spectral Clustering with Spatial Regularity Experimental Evaluation Dataset and Evaluation Method Results Conclusions References Spotlights and Posters R1 Exploiting Repetitive Object Patterns for Model Compression and Completion Introduction Related Work Overview Extraction of Mutually Similar Object Instances Analysis of Repetitive Objects Latticelets Cycles and Chains Structure Inference Using Conditional Random Fields Conditional Random Fields Node and Edge Features Network Structure Model Compression Experiments Conclusions References Feature Tracking for Wide-Baseline Image Retrieval Introduction Related Work Bag-of-Visual-Words Image Retrieval Constructing the $Track-Graph$ Feature Matching and Track Extraction Graph Construction Properties of the Track-Graph Image Retrieval with the Track-Graph Evaluation Evaluation Criteria Results Future Work References Crowd Detection with a Multiview Sampler Introduction Related Work A Gibbs Point Process for Crowd Detection Modeling Inference Multiview Proposals Experiments Conclusion References A Unified Contour-Pixel Model for Figure-Ground Segmentation Introduction Related Work Localization and Segmentation Models Outline Localization Foreground Segmentation Image-Specific Appearance TheContour-PixelModel Superpixel-Based Inference Inference Challenges Joint Inference Experimental Results Discussion References SuperParsing: Scalable Nonparametric Image Parsing with Superpixels Introduction System Description Retrieval Set Superpixel Features Local Superpixel Labeling Contextual Inference Simultaneous Classification of Semantic and Geometric Classes Results Large Datasets Small Datasets Discussion References Segmenting Salient Objects from Images and Videos Introduction Saliency Measure Definition of Saliency Measure Regularized Saliency Measure Implementation Salient Object Segmentation Segmentation Energy for Still Images Segmentation Energy for Videos Experiments Segmenting Salient Objects from Images Segmenting Salient Objects from Video Sequences Conclusions References ClassCut for Unsupervised Class Segmentation Introduction OverviewofOurMethod Segmentation Prior $ΦΘ(L, I)$ Class Model $ΨΘ(L, I)$ Energy Minimization Initializing and Updating the Class Model Location Model Shape Model Appearance Model Finding the Reference Frame Experiments Datasets Baselines and the State of the Art ClassCut Conclusion References A Dynamic Programming Approach to Reconstructing Building Interiors Introduction Background Outline of Proposed Approach Identifying Dominant Directions Identifying the Floor and Ceiling Planes Rectifying Vertical Lines Obtaining Weak Orientation Estimates Formulation of Reconstruction Problem Formalisation Proposed Algorithm Auxiliary Sub–problems From ($L^3K$) to $O$($L^2K$) Results Failure Cases Discussion References Discriminative Mixture-of-Templates for Viewpoint Classification Introduction Related Work Discrete Viewpoint Models Training Supervised Case Semi-supervised Case Unsupervised Case Continuous Viewpoint Models Training Experimental Evaluation: Discrete Viewpoints Supervised Case Semi-supervised Case Unsupervised Case Experimental Evaluation: Continuous Viewpoints Conclusion References Efficient Non-consecutive Feature Tracking for Structure-from-Motion Introduction Related Work Our Approach Two-Pass Matching for Consecutive Tracking First-Pass Matching by Descriptor Comparison Second-Pass Matching by Planar Motion Segmentation Non-consecutive Track Matching Fast Matching Matrix Estimation Non-consecutive Track Matching Tracks in Multiple Videos Results Conclusion and Discussion References $P2Π$: A Minimal Solution for Registration of 3D Points to 3D Planes Introduction and Previous Work Pose Estimation The Choice of Intermediate Coordinate Frames The Use of Coplanarity Constraints The Use of Orthonormality Constraints Other Variants The Correspondence Problem A General Framework for Pose Estimation Experimental Results Discussion References Boosting Chamfer Matching by Learning Chamfer Distance Normalization Introduction Related Work Oriented Chamfer Distance (OCD) Normalization of Oriented Chamfer Distance Learning Normalized OCD with AdaBoost Object Detection with NOCD Normalizers Experimental Evaluation of Detection Rate Conclusions References Geometry Construction from Caustic Images Introduction Related Work Optimization Framework Optimization Using SPSA Caustic Rendering on the GPU Results Discussion Conclusion References Archive Film Restoration Based on Spatiotemporal Random Walks Introduction Background Proposed Method Preliminaries and Definitions Restoration of Degraded Pixels Multiscale Refinement Experimental Results and Discussion Conclusion References Reweighted Random Walks for Graph Matching Introduction Problem Formulation Random Walks for Graph Matching Affinity-Preserving Random Walks Reweighted Random Walks Experiments Synthetic Random Graph Matching Feature Point Matching across Image Sequences Real Image Matching Conclusion References Rotation Invariant Non-rigid Shape Matching in Cluttered Scenes Introduction Shape Representation Outlier Resistant Shape Context Distance Energy Function Algorithm Experimental Results Experiments Using Synthetic Data Experiments on Real Data Conclusion References Loosely Distinctive Features for Robust Surface Alignment Introduction Game-Theoretic Matching Surface Hashes Interest Points Detection Matching Surface Hashes Experimental Results Sensitivity to Noise, Occlusion, and Scale of the Descriptor Comparisons with Other Matchers System-Level Comparisons Conclusions References Accelerated Hypothesis Generation for Multi-structure Robust Fitting Introduction Related Work Inlier Probabilities from Sorting Information Guided Sampling for Multi-structure Data Experiments Multiple Line and Circle Fitting Homography Estimation Fundamental Matrix Estimation Conclusions References Aligning Spatio-Temporal Signals on a Special Manifold Introduction The Framework of Alignment Problem The Alignment Manifold The Spatial Alignment Submanifold The Temporal Alignment Submanifold Sequential Importance Sampling on the Manifold for Optimal Alignment Empirical Evaluation Evaluation with Point Trajectories Evaluation with Deforming Shape Sequences Evaluation with Human Action Videos Discussion References Supervised Label Transfer for Semantic Segmentation of Street Scenes Introduction Overview KNN-MRF Matching Superpixel Descriptor Distance Metric Matching Correspondences Classification Confidence Inference and MRF Integration Selection of Proper Image Sets for Label Transfer Confidence Inference Influence of Matching Correspondences Classification Markov Random Field Integration Experiments Google Street View Dataset CBCL StreetScenes Dataset CamVid Dataset Computation Time Conclusion References Category Independent Object Proposals Introduction Related Work Overview of Approach Proposing Regions Hierarchical Segmentation Seeding Generating Segmentations Ranking Proposals Experiments and Results Proposal Generation Ranking Performance Discussion References Photo-Consistent Planar Patches from Unstructured Cloud of Points Introduction Overview of the J-linkage Algorithm Random Sampling Agglomerative Clustering Constraints Integration Photo-Consistency Constraint Visibility Constraint Non Intersection Constraint Filling the Gaps Results Comparative Test Real World Examples Discussion References Contour Grouping and Abstraction Using Simple Part Models Introduction Related Work Overview of the Approach Finding Consistent Cycles Path Initialization Path Extension Training the Classifiers Abstracting the Shape of a Consistent Cycle Results Conclusions References Dynamic Color Flow: A Motion-Adaptive Color Model for Object Segmentation in Video Introduction Related Work Dynamic Color Flow DCF for Video Object Segmentation The Foreground Layer The Background Layer Segmentation with Shape Priors Experiments and Comparisons Concluding Remarks References What Is the Chance of Happening: A New Way to Predict Where People Look Introduction Motivation From Low Level Feature Priors to $CoH$ Single Feature Probability Distribution Surrounding Feature Prior Estimation of $CoH$ CenterBiasPrior Computational Framework for Saliency Experiments Qualitative Evaluation Quantitative Evaluation Conclusions References Supervised and Unsupervised Clustering with Probabilistic Shift Introduction Related Work Approach Detection of Isotropic Density Neighborhoods Shift Vector Computation Cluster Identification Algorithms Partitioning by Connected Destinations Supervised and Unsupervised Spectral Partitioning Results Artificial Data Real Data Conclusions and Contributions References Depth-Encoded Hough Voting for Joint Object Detection and Shape Recovery Introduction Previous Work Depth-Encoded Hough Voting Training the Model Recognition and 3D Reconstruction Evaluation Exp.I: System Analysis on a Novel 3D Table-Top Object Dataset Exp.II:Comparision on Three Challenging Datasets Applications: 6 DOF Pose Estimation and 3D Object Modeling Conclusion References Shape Analysis of Planar Objects with Arbitrary Topologies Using Conformal Geometry Introduction Theoretically Background Beltrami Equation Conformal Module Holomorphic Differentials Conformal Welding Algorithm Shape Signatures of Planar Domains with Arbitrary Topologies Reconstruction of Shapes from Signatures Experimental Results Conclusion and Future Work References A Coarse-to-Fine Taxonomy of Constellations for Fast Multi-class Object Detection Introduction Related Work Overview and Contributions Representation: Coarse-to-Fine Constellation Taxonomy The Probabilistic Constellation Model of Objects A Coarse-to-Fine Taxonomic Constellation Tree (TCT) EfficientInference Learning the Taxonomic Constellation Tree Combining TCT with the Hierarchy-of-Parts Model [4] Experimental Results ETHZ Shape Dataset: 5 Object Classes TUD Shape Dataset: 10 Object Classes Shape 15 Caltech 101 and LabelMe Dataset (148 Object Classes) Summary and Conclusions References Object Classification Using Heterogeneous Co-occurrence Features Introduction Co-occurrence Histograms of Oriented Gradients Proposed Features Color-CoHOG CoHED CoHD Color Histogram Experiment 1. INRIA Person Dataset Feature Evaluation Comparison with CoHOG Comparison with Previous Methods Experiment 2. Oxford 17/102 Category Flower Datasets Conclusion References Converting Level Set Gradients to Shape Gradients Introduction The Family of Admissible SDF Variations Velocity Projection Gradients and Inner Products Relating the Two Gradients The Correct Way to Evolve the Level Sets Implementation Implications for Common Level Set Gradients Discussion and Conclusions References A Close-Form Iterative Algorithm for Depth Inferring from a Single Image Introduction Related Works Overview of Our Algorithm The Framework of Our Algorithm Multi-scale Image Segmentation Depth Inferring Region Merging Experiments Scene Reconstruction with Fore-Object Conclusion References Learning Shape Segmentation Using Constrained Spectral Clustering and Probabilistic Label Transfer Introduction Paper Contributions Laplacian Embeddings and Their Properties Propagating Pairwise Constraints Shape Segmentation via Label Transfer Experiments and Results Conclusions References Weakly Supervised Shape Based Object Detection with Particle Filter Introduction Related Work Partially-Supervised Model Learning Part Model Construction Relation between Model Parts Framework for Object Detection Evaluation Based on Shape Similarity Experimental Results Detection according to Bounding Boxes Localizing Object Boundaries Conclusion and Discussion References Geodesic Shape Retrieval via Optimal Mass Transport Introduction Feature-Based Shape Retrieval Contributions Geodesic Distances Geodesic Distance Definition Geodesic Distance Computation Geodesic Descriptors Local Descriptors Global Descriptors Optimal Transport Retrieval Similarity Measure Wasserstein Distance Approximate Wasserstein Distance Numerical Examples 2-D Shape Retrieval 3-D Shape Retrieval Conclusion References Spotlights and Posters R2 Image Segmentation with Topic Random Field Introduction Preliminary: Image Representation Topic Random Field Spatial MRF over Topic Assignments Noise Channel over Codebook The Proposed Model Variational Inference and Parameter Learning Experiments Data Sets Experimental Setups and Comparisons Image Segmentation Results Conclusions References Author Index Annotation The six-volume set comprising LNCS volumes 6311 until 6313 constitutes the refereed proceedings of the 11th European Conference on Computer Vision, ECCV 2010, held in Heraklion, Crete, Greece, in September 2010. The 325 revised papers presented were carefully reviewed and selected from 1174 submissions. The papers are organized in topical sections on object and scene recognition; segmentation and grouping; face, gesture, biometrics; motion and tracking; statistical models and visual learning; matching, registration, alignment; computational imaging; multi-view geometry; image features; video and event characterization; shape representation and recognition; stereo; reflectance, illumination, color; medical image analysis The six-volume set comprising LNCS volumes 6311 -- 6316 constitutes the refereed proceedings of the 11th European Conference on Computer Vision, ECCV 2010, held in Heraklion, Crete, Greece, in September 2010. The 325 revised papers presented were carefully reviewed and selected from 1174 submissions. The papers are organized in topical sections on object and scene recognition; segmentation and grouping; face, gesture, biometrics; motion and tracking; statistical models and visual learning; matching, registration, alignment; computational imaging; mutli-view geometry; image features; video and event characterization; shape representation and recognition; stereo; reflectance, illumination, color; medical image analysis