Computer Vision -- ACCV 2009: 9th Asian Conference on Computer Vision, Xi'an, China, September 23-27, 2009, Revised Selected Papers, Part II (Lecture Notes in Computer Science, 5995)

It givesus greatpleasureto presentthe proceedings of the 9th Asian Conference on Computer Vision (ACCV 2009), held in Xi’an, China, in September 2009. This was the ?rst ACCV conference to take place in mainland China. We received a total of 670 full submissions, which is a new record in the ACCV series. Overall, 35 papers were selected for oral presentation and 131 as posters, yielding acceptance rates of 5.2% for oral, 19.6% for poster, and 24.8% in total. In the paper reviewing, we continued the tradition of previous ACCVsbyconductingtheprocessinadouble-blindmanner.Eachofthe33Area Chairs received a pool of about 20 papers and nominated a number of potential reviewers for each paper. Then, Program Committee Chairs allocated at least three reviewers to each paper, taking into consideration any con?icts of interest and the balance of loads. Once the reviews were ?nished, the Area Chairs made summaryreportsforthepapersintheirpools,basedonthereviewers’comments and on their own assessments of the papers. Title Page Preface Organization Table of Contents Poster Session 1: Stereo, Motion Analysis, and Tracking A Dynamic Programming Approach to Maximizing Tracks for Structure from Motion Introduction Related Work The Subtrack Optimization Algorithm Terminology and Problem Definition A Dynamic Programming Solution The Complete Structure from Motion Process Performance Considerations Results Conclusion References Dense and Accurate Spatio-temporal Multi-view Stereovision Introduction Our Approach Discretize Then Optimize Animated Mesh Representation Variational Formulation Data Attachment Term Velocity Field Regularization Term Numerical Experiments Implementation Aspects Experimental Results Conclusion References Semi-supervised Feature Selection for Gender Classification Introduction PGA on Facial Needle-Maps Learning Gender Discriminating Features Semi-supervised Learning Using Harmonic Functions Application to Gender Classification Experimental Results Conclusions References Planar Scene Modeling from Quasiconvex Subproblems Introduction and Related Work SVD Based Initializations Optimization Framework Formulation of the Objective Function Universal Scale Retrieving Rotation Experimental Analysis Synthetic Data Real Data Discussion and Conclusion References Fast Depth Map Compression and Meshing with Compressed Tritree Introduction Related Work Proposed Scheme Content Adaptive Meshing with Tritree Residual Coding Results Conclusion References A Three-Phase Approach to Photometric Calibration for Multi-projector Display Using LCD Projectors Introduction Three-Phase Photometric Model Generalized Color Matching Process Single-Projector Model Normalized-Projector Model Display-Wall Model Photometric Calibration Algorithm Experimental Results Setup Result Conclusions and Future Work References Twisted Cubic: Degeneracy Degree and Relationship with General Degeneracy Introduction Preliminaries Degeneracies from Twisted Cubic Degeneracy Degree from Twisted Cubic Relationship with Ruled Quadric Degeneracy Relationship with H-degeneracy Experiment Simulations Experiments on Real Data Conclusion References Two-View Geometry and Reconstruction under Quasi-perspective Projection Introduction Camera Projection Geometry Two-View Geometry and Reconstruction Fundamental Matrix Plane Induced Homography RANSAC Algorithm with Outliers 3D Structure Reconstruction Evaluation on Synthetic Data Evaluation on Real Images Conclusion References Similarity Scores Based on Background Samples Introduction Related Work The Two-Shot Similarity Score Background-Sample Based Similarities with LDA Ranking Based Background Similarity Face Alignment Experiments The Contribution of Alignment The Contribution of One-Shot The Contribution of Two-Shot The Contribution of the Ranking Descriptor Combining Background Similarities beyond LDA Conclusions References Human Action Recognition Using Spatio-temporal Classification Introduction Learning for Spatio-temporal Classification Spatial Subspace Creation Using Locality Preserving Projection Learning for Classification in Temporal Subspace Recognition Process Experimental Results Human Action Databases Recognition Results and Comparison Performance Evaluation in Noisy Data Conclusions References Face Alignment Using Boosting and Evolutionary Search Introduction Boosting Appearance Models Shape and Texture Models in Active Appearance Models [2] AppearanceModeling Real Adaboost Learning for Strong Classifier Learning Sparse Granular Features forWeak Classifier Granular Features Bayesian Stump Look Up Table Weak Classifier Evolutionary Search for Sparse Feature Selection Face Fitting with Boosting Classifier Experiments Conclusion References Tracking Eye Gaze under Coordinated Head Rotations with an Ordinary Camera Introduction Eye-Head Coordination Study Tracking Head-Coordinated Eye Gaze The Problem Formulation Deriving the Compensation Function $\sqsupset$ Eye Gaze Tracking Framework Extracting Iris-Corner Vector 3D Head Pose Tracker Extracting the Head-Normalized Gaze Vector Experimental Results Conclusion References Orientation and Scale Invariant Kernel-Based Object Tracking with Probabilistic Emphasizing Introduction Mean Shift Tracking: Brief Review The Proposed Method Probabilistic Emphasizing Orientation Estimation Scale Adaptation Algorithm Summary Experiments Conclusion References Combining Edge and Color Features for Tracking Partially Occluded Humans Introduction Related Work Proposed Approach Appearance Model Mean Shift Part Tracking Edge Matching Part Assignment and Target Localization Results Conclusion and Future Work References Incremental Multi-view Face Tracking Based on General View Manifold Introduction Model Description of Incremental Multi-view Face Tracking View Estimation Based on the General View Manifold The View Manifold Generation Method View Estimation Based on Point-to-Manifold Distance Immediately Emphasized Online Updating Experimental Results and Analysis Conclusions and Future Work References Hierarchical Model for Joint Detection and Tracking of Multi-target Introduction RelatedWork Hierarchical Composition Model And-Or Graph for MTTV Inference Algorithm Top-Down Channel The State Prediction Distribution Posteriors of Association and Existence Monte Carlo Data Association Filtering Context Sensitive Reasoning Channel Bottom-Up Channel and Hybrid Measurement Process Experiments and Discussion Performance Evaluation Tracking Football Players and Pedestrians Conclusion References Heavy-Tailed Model for Visual Tracking via Robust Subspace Learning Introduction Overview of Previous Work Subspace Learning Method for Tracking The Extended and Improved Incremental PCA Robust PCA Bayesian Inference for Visual Tracking Heavy-Tailed Dynamical Model IRPCA-Based Observation Model Implementation and Experiments Experimental Results Qualitative Comparison Conclusions and Future Work References Efficient Scale-Space Spatiotemporal Saliency Tracking for Distortion-Free Video Retargeting Introduction Distortion-Free Video Retargeting Problem Formulation Video Information Loss Constraints for Video Retargeting Detecting and Tracking Salient Regions Scale-Space Spatiotemporal Saliency Experiments Conclusion and Future Work References Visual Saliency Based Object Tracking Introduction Problem Description and Modeling The Saliency Features and Constraints Static Saliency Features Motion Saliency Features Feature Selection and Combination Coherence Constraints Particle Filter Tracking Experiments Color Spatial Distribution Tracking Results Conclusion References People Tracking and Segmentation Using Efficient Shape Sequences Matching Introduction Previous Work Computing Shape Priors Matching Measure $O$(1) Tanimoto Distance Computation Embedding Shape Priors in Min-Cut Segmentation Embedding Shape Priors Experimental Results Refinement of Bounding Boxes and Phase Estimation Segmentation Results Quantitative Evaluation Conclusions References Monocular Template-Based Tracking of Inextensible Deformable Surfaces under $L_{2}$-Norm Introduction Deformable Surface 3D Tracking under $L_{2}$-Norm Problem Statement $L_{2}$-Norm Approximation Shape Regularization Experimental Results Synthetic Data Real Data Conclusion References A Graph-Based Feature Combination Approach to Object Tracking Introduction Kernel-Based Tracking in Arbitrary Feature Spaces Feature Combination via Graph Embedding Feature Mapping Double Centering Minimising Feature Correlation On-Line Feature Weight Updating Algorithm Description Experiments Conclusion References A Smarter Particle Filter Introduction Particle Filter and Its Limitation Particle Filter Limitation Annealed Gaussian Based PSO Traditional PSO Annealed Gaussian Based PSO Swarm Intelligence Based Particle Filter Motivation The Smarter Particle Filter Theoretical Analysis from Bayesian Filtering View Experimental Results State Estimation Visual Tracking Conclusion References Robust Real-Time Multiple Target Tracking Introduction Related Work Rao-Blackwellized Resampling Particle Filter with Fixed-Lag Experimental Results Tracking Soccer Players Tracking Ants Conclusions References Dynamic Kernel-Based Progressive Particle Filter for 3D Human Motion Tracking Introduction 3D Human Model Observation Likelihood Estimation Tracking Approach Particle Filter Progressive Particle Filter Dynamic Kernel Model Dynamic Kernel Based Progressive Particle Filter Experimental Results Conclusion References Bayesian 3D Human Body Pose Tracking from Depth Image Sequences Introduction Human Body Model and Pose Estimation with Constraint Inverse Kinematics Robust 3D Pose Tracking with Bayesian Method Key-Point Detection from Depth Image Sequence for Pose Tracking Temporal Prediction, Density Sampling and Dense Correspondence Searching for Pose Tracking Tracking Error Evaluation Bayesian Updating and MAP Selection Experiments Conclusion References Crowd Flow Characterization with Optimal Control Theory Introduction Related Work Estimation by Coupling Observations/Model with Variational Assimilation Overview of the Method Variational Assimilation Dynamic Model, Observations and Covariance Experimentations and Discussion Validation Experimention on Real Crowd Discussion Conclusion and Future Works References Human Action Recognition Using HDP by Integrating Motion and Location Information Introduction BasicIdea Features Motion Feature Location Information Hierarchical Dirichlet Processes - Latent Dirichlet Allocation Multiple Instances HDP-LDA Experiments Conclusion References Detecting Spatiotemporal Structure Boundaries: Beyond Motion Discontinuities Introduction Technical Approach Spatiotemporal Oriented Energy Representation Anisotropic Smoothing Spatiotemporal Structure Boundaries Algorithm Empirical Evaluation Discussion and Summary References An Accelerated Human Motion Tracking System Based on Voxel Reconstruction under Complex Environments Introduction Accelerated Robust Reconstruction of Human Voxels The Fusion Framework Using Multiple Camera Information on the Pixel Level Combine Pixel and Super-Pixel Classification under Complex Environment Reconstruction Acceleration by Focusing on Region of Interest Human Motion Tracking Using Hierarchical Optimization Skeleton Model and Automated Tracking Initialization Tracking by Hierarchical Optimization Experimental Results and Quantitative Evaluation Reconstruction Results and Evaluation Tracking Results and Evaluation Conclusions and Future Work References Automated Center of Radial Distortion Estimation, Using Active Targets Introduction ActiveTarget Center of Radial Distortion Estimation Hartley’s Method Active Center of Distortion Estimation Active Approach Implementation Experiments Conclusion References Rotation Averaging with Application to Camera-Rig Calibration Introduction Existing Works on Rotation Averaging Problem Formulation Averaging Rotations The Geodesic $L_{2}$-Mean The Geodesic $L_{1}$-Mean Quaternion Averaging The Conjugate Averaging Problem Iterative Rotation Averaging for Camera Rig Calibration Experiments Synthetic Rotation Averaging Synthetic Camera Rig Experiments on Real Images Conclusion and Future Work References Single-Camera Multi-baseline Stereo Using Fish-Eye Lens and Mirrors Introduction Proposed System System Configuration Equivalent Multi-Camera System Dimensional Design of Mirrors System Calibration Unified Projection Model for a Fish-Eye Lens Lens Calibration Position and Orientation of the Mirrored Cameras Single-Camera Multi-baseline Stereo Perspective Reprojection of the Fish-Eye Image Matching in the Fish-Eye Image Experimental Results Conclusions References Generation of an Omnidirectional Video without Invisible Areas Using Image Inpainting Introduction Generation of an Omnidirectional Video without Invisible Areas Estimation of Extrinsic Camera Parameters and Positions of Natural Feature Points Generation of Images Projected on Planes by Estimating Shapes Around Missing Regions Inpainting a Missing Region Based on Energy Minimization Generation of an Omnidirectional Video Using Inpainted Images Experiments Inpainting a Missing Region in an Omnidirectional Video Omnidirectional Telepresence without Invisible Areas Conclusion References Accurate and Efficient Cost Aggregation Strategy for Stereo Correspondence Based on Approximated Joint Bilateral Filtering Introduction and Related Work Proposed Cost Aggregation Strategy Experimental Results Conclusions References Detecting Critical Configurations for Dividing Long Image Sequences for Factorization-Based 3-D Scene Reconstruction Introduction Factorization-Based 3-D Reconstruction Notations Matrix Rank-$r$ Factorization Rank-$r$ Iteration: Projective Reconstruction Using Separate Bundle Adjustment Factorization-Based Self-calibration The Proposed Algorithms Algorithm for Counting Distinct Camera Centers (ACCC) Algorithm for Detecting Critical Configurations (ADCC) Discussion on Detection of Critical Configurations Algorithm for Dividing Long Image Sequence (ADLS) Experimental Results Detecting Pure Rotation and Coplanar 3-D Points Counting Distinct Camera Centers Dividing Long Image Sequences Conclusion References Scene Gist: A Holistic Generative Model of Natural Image Introduction Related Works Scene Gist Generative Model Information Redundancy Revisited Learning Gist Component Discriminative Property of Gist Components Experiment Gist Subspace Classification Method Result Discussion and Conclusion References A Robust Algorithm for Color Correction between Two Stereo Images Introduction Methodology OF-SIFT Experiments Performance of OF-SIFT Performance of Color Correction Algorithm Experiments under Extreme Conditions Experiments on Real World Scene Conclusions References Efficient Human Action Detection Using a Transferable Distance Function Introduction Human Action Detection Motion Feature Patch Based Action Comparison Cascade Structure Transferable Distance Function Construction of Cascade Structure Experiments Datasets Experiments on the Cluttered Action Dataset Experiment on the Ballet Video Conclusion References Crease Detection on Noisy Meshes via Probabilistic Scale Selection Introduction Multi-scale Representation of a Triangular Mesh Anisotropic Diffusion Probabilistic Estimation of Local Scale Minimum Description Length Scale Selection with a Prior MRF Model Multi-scale Approach to Crease Detection Experiments Conclusions References Improved Uncalibrated View Synthesis by Extended Positioning of Virtual Cameras and Image Quality Optimization Introduction Uncalibrated View Synthesis Extended Positioning of Virtual Cameras Camera Movement along the Line of Sight Camera Movement Orthogonally to the Geodesic and the Line of Sight Image Quality Optimization (Utilize Information of Both Reference Images) Experimental Results Conclusion References Region Based Color Image Retrieval Using Curvelet Transform Introduction The Curvelet Transform Curvelet Feature Extraction for Irregular Region Irregular to Regular Shape Transform Feature Extraction and Distance Measure Retrieval Performance Conclusion References Extracting Spatio-temporal Local Features Considering Consecutiveness of Motions Introduction Related Work Proposed Method Extraction of Appearance Features Selection of Motion Points Extraction of Motion Features Generation of Local Spatio-temporal Features Experimental Results Action Recognition Web Video Shot Classification Conclusion References Multi-view Texturing of Imprecise Mesh Introduction Morphing Images to Adapt to the Mesh Overview of the Algorithm Interest Point Matching Reprojection of 3D Points through the Mesh Dense Deformation Texture Mapping Experiments Conclusion References Poster Session 2: Segmentation, Detection, Color and Texture Semantic Classification in Aerial Imagery by Integrating Appearance and Height Information Introduction Semantic Classification Approximated Covariance Representation Randomized Forest Framework Incorporating Contextual Information Experimental Evaluation Experiments on MSRC-9 Dataset Experiments on Aerial Images Conclusion References Real-Time Video Matting Based on Bilayer Segmentation Introduction Segmentation Algorithm Framework Color Likelihood Propagation Matting Algorithm Bayesian Matting Proposed Method Experimental Results Evaluation of Segmentation Evaluation of Matting Calculation Time Conclusions References Transductive Segmentation of Textured Meshes Introduction Graph Laplacian Transduction Transductive Inference Graph Laplacian Regularization and Energy Minimization Multi-class Segmentation Segmentation of Textured Meshes Our Algorithm Features and Kernels Additional Knowledge on the Test Points Computational Complexity and Time Experimental Results Conclusion References Levels of Details for Gaussian Mixture Models Introduction and Prior Work Entropic Quantization of GMMs Relative Entropy and Bregman Divergence Bregman $k$-Means Symmetric Bregman $k$-Means Hierarchical GMM Representation Experiments Bregman $k$-Means Clustering Method Comparison BKMC versus UTAC Clustering-Based Image Segmentation Hierarchical GMM Representation Concluding Remarks References A Blind Robust Watermarking Scheme Based on ICA and Image Dividing Blocks Introduction Watermarking Using ICA Watermark Embedding Scheme Watermark Extraction Scheme Estimate Watermark Based on ICA The Post-processing Scheme Experimental Performance Analysis Simulate Setup JPEG Compression Test Gray Scale Reduction Test Other Attacks Authentication Test Conclusion References MIFT: A Mirror Reflection Invariant Feature Descriptor Introduction MIFT Mirror Reflection Descriptor Reorganization Matching Results Applications Application to Image Search Application to Detection of Symmetry Axis Conclusion References Detection of Vehicle Manufacture Logos Using Contextual Information Introduction Related Work Vehicle Logo Detection Based on Contextual Information License Plate Localization Headlights Localization Contextual Information from a License Plate and Headlights Contextual Information Using Grille Experimental Results Conclusions References Part-Based Object Detection Using Cascades of Boosted Classifiers Introduction Model Integral Histogram of HOG and Gradient Features Cascades of Boosted Classifiers Spatial Relations Learning Training the Cascade Training Spatial Relations Detection and Localization Experiments Conclusion References A Novel Self-created Tree Structure Based Multi-view Face Detection Introduction Kernel-Based Branching Competitive Learning and Multi resolutionClustering Tree Branching Competitive Learning Network Kernel-Based BCL in Reduced Empirical Feature Space Kernel BCL Based Multi-resolution Clustering Tree Support Vector Machines and Boosted MPP Classifiers Experimental Results On Single SVM and the Boosted MPP Classifiers On Multi-view Face Detection Conclusions References Multilinear Nonparametric Feature Analysis Introduction Related Work Nonparametric Discriminant Analysis Nonparametric Feature Analysis Multilinear Nonparametric Feature Analysis (MNFA) Multilinear Algebra and Notations Optimal Criterion of MNFA Optimization of $k$-Mode Optimal Solution of MNFA Classification with MNFA Experiments Conclusions References A Harris-Like Scale Invariant Feature Detector Introduction Feature Model and Eigenvalue Description Our Method: HLSIFD Detector Procedure Matching and Description Procedure Experimental Results Conclusions References Probabilistic Cascade Random Fields for Man-Made Structure Detection Introduction Review of Discriminative Random Fields Probabilistic AdaBoost Cascade Probabilistic Cascade Random Fields Experiment Features Results Conclusions and Future Works References A Novel System for Robust Text Location and Recognition of Book Covers Introduction Robust Feature Extraction Text Location Adaptive Binarization Kasar’s Method [16] Evaluation of the Original Algorithm Improved Kasar’s Method Experiments Conclusion References A Multi-scale Bilateral Structure Tensor Based Corner Detector Introduction Harris Corner Detector Bilateral Structure Tensor Based Corner Detection Construction of the Bilateral Structure Tensor Multi-scale Filtering Experimental Results Conclusions References Pedestrian Recognition Using Second-Order HOG Feature Introduction Related Works Second-Order Histograms of Oriented Gradients Histogram of Gradient Orientations Pairwise Relationships among Histogram Components Experimental Results Conclusion References Fabric Defect Detection and Classification Using Gabor Filters and Gaussian Mixture Model Introduction Defect Detection Using Gabor Filters Improved Gabor Filters Optimal Output Selection Binarization Defect Classification Using GMM Feature Extraction GMM Based Classification Experiment and Analysis Defect Detection Defect Classification Comparison with Other Techniques for Fabric Defect Classification Conclusions References Moving Object Segmentation in the H.264 Compressed Domain Introduction Related Works Moving Object Segmentation in H.264 Compressed Domain Global Motion Estimation Motion Vector Similarity Measure MV Field Spatio-temporal Refinement Object Detection in Intra Frame Refinement in Non-intra Frame Experiment and Result Conclusions References Video Segmentation Using Iterated Graph Cuts Based on Spatio-temporal Volumes Introduction Graph Cuts for Video Segmentation Graph Cuts for Image Segmentation Problems with Conventional Method Iterated Graph Cuts Using Spatio-temporal Volumes Proposed Method Spatio-temporal Volume Add Edges Using Corresponding Points Iterated Graph Cuts Experimental Results Experiment Outline Comparison with Conventional Method Comparison of Graph Size Effect of Adding Edge Using Corresponding Points Conclusion References Spectral Graph Partitioning Based on a Random Walk Diffusion Similarity Measure Introduction Spectral Graph Partitioning Solving K-Way Normalized Cuts Random Walk Diffusion Similarity Measure Our Defined One-Step RandomWalk on a Graph Our Generalized $m$-Step RandomWalk on a Graph Random Walk Diffusion Similarity Measure Our Spectral Graph Partitioning Objective Function Finding Optimal Solutions Experiments Conclusion References Iterated Graph Cuts for Image Segmentation Introduction Image Segmentation by Graph Cuts The Iterated Graph Cuts Iterated Conditional Mode The Iterated Graph Cuts Algorithm Experimental Results Comparison with Standard Graph Cuts Comparison with GrabCut Quantitative Evaluation Discussion Conclusion and Future Work References Contour Extraction Based on Surround Inhibition and Contour Grouping Introduction Surround Inhibition Process Original Computational Method Improved Scheme Thinning and Binarization Contour Group Local Interactions of Contour Elements Global Contour Perceptions Experiment Experimental Results Conclusion References Confidence-Based Color Modeling for Online Video Segmentation Introduction Confidence-Based Segmentation Model Confidence-Based Color Modeling (CCM) Global Color Model Per-pixel Color Model Optimal Combination Experimental Results Conclusions References Multicue Graph Mincut for Image Segmentation Introduction Multicue Graph Mincut Framework Multicue Embedded Gibbs Energy Self-validated Segmentation Experimental Results Conclusions References Author Index Annotation The three volume set LNCS 5994, LNCS 5995, and LNCS 5996 constitutes the thoroughly refereed post-conference proceedings of the 9th Asian Conference on Computer Vision, ACCV 2009, held in Xi'an, China, inSeptember 2009. The 35 revised full papers and 130 revised poster papers of the three volumes were carefully reviewed and seleceted from 670 submissions. The papers are organized in topical sections on multiple view and stereo, face and pose analysis, motion analysis and tracking, segmentation, feature extraction and object detection, image enhancement and visual attention, machine learning algorithms for vision, object categorization and face recognition, biometrics and surveillance, stereo, motion analysis, and tracking, segmentation, detection, color and texture, as well as machine learning, recognition, biometrics and surveillance