位置：立项数据库 > 立项详情页
快速鲁棒迁移子空间算法
  • 项目名称：快速鲁棒迁移子空间算法
  • 项目类别：面上项目
  • 批准号：61072093
  • 申请代码：F010401
  • 项目来源：国家自然科学基金
  • 研究期限：2011-01-01-2013-12-31
  • 项目负责人：李学龙
  • 负责人职称：研究员
  • 依托单位：中国科学院西安光学精密机械研究所
  • 批准年度：2010
中文摘要：

子空间分析在图像处理和模式识别中重要而应用广泛。现有子空间分析方法均假设训练数据和测试数据独立同分布。但实际情况中，训练样本和测试样本常来自不同领域，独立同分布的假设并不成立。因而现有方法对训练数据过拟合严重，而对测试样本的泛化能力急剧下降。为解决非独立同分布带来的问题，本项目把迁移学习(transfer learning)思想引入到子空间分析中。对概率子空间分析和常规（指基于似然函数最大化和基于鉴别信息保持的方法）子空间分析，本项目分别研究统一理论框架，使目标函数中体现训练数据的知识向测试数据知识的迁移，从而使子空间分析能最大程度地适应非独立同分布的情况（称为迁移性）。称其为迁移子空间分析。本项目还研究如何提高迁移子空间分析对离群点的鲁棒性，以及研究如何增强迁移子空间分析在特征提取阶段的实时性。总之，本项目旨在研究具有迁移性、鲁棒性和实时性的迁移子空间分析，促进其在图像处理和模识中应用。

中文主题词： 子空间学习；维数约简；迁移学习；模式识别；图像处理
英文摘要：

subspace learning；dimensionality reduction；transfer leanring；pattern recognition；image processing

英文主题词： subspace learning；dimensionality reduction；transfer leanring；pattern recognition；image processing
结论摘要：

在子空间学习、迁移学习及其在视觉计算中的应用等方面发表论文92篇，其中SCI检索59篇，T-PAMI等IEEE汇刊论文39 篇，IJCV论文1篇，CVPR5篇在内的会议论文33 篇。在IEEE TNNLS汇刊上一篇迁移学习的综述已经Accepted with Minor Revision。项目总体主要科研成果1. 提出了基于Bregman dirvergence和迁移学习的潜在变量模型。基于潜在变量模型的子空间学习算法假设训练和测试样本满足独立同分布，故在训练集合上得到模型很难适用于测试集。为解决该问题，提出一种参数迁移学习的潜在变量模型，无需针对测试集合重新建立模型，而只需根据训练集和测试集之间的Bregman divergence进行参数调整即可得到潜在变量模型，使其适应于不同的数据集。2. 提出了基于迁移学习的行人检测。用不同场景数据集共享的相似分布，提出利用训练数据集中的数据来构建新场景中的行人检测模型。对训练场景中的错分样本和未知场景的错分样本采用不同的权重调整策略，得到具备迁移学习特性的行人检测。3. 提出基于随机游走的直推式半监督子空间方法。传统半监督方法在传播类别标签时，当无标签数据与有标签数据差异较大时会产生错误传播。为克服该问题，提出基于随机游走的直推式半监督子空间方法，用加权回归模型对随机游走得到虚拟标签进行编码，在随机游走时能够检测离群点，减少错误的标签传播。4. 提出基于截断核范数正则化的快速、准确的矩阵完整化方法。现有方法在从少数元素已知的信息中恢复大矩阵时，将问题表示成一个低秩逼近问题，用核范数作为凸松弛，需同时最小化所有奇异值，故在实际中难很好近似秩。为解决此，提出基于截断核范数的矩阵秩的逼近方法，用最大的若干个奇异值减去核范数。在此基础上，提出一种矩阵完整化方法，且提出了三种有效的迭代算法。5. 提出基于多模式的图像概率子空间表示模型。对高维数据，小样本问题严重。提出一种基于概率主成分分析模型的子空间学习。显式表示图像变换机制并且将其纳入模型的随机成分中。通过模型自动拟合可学习到最佳的变换。6. 提出了基于联合嵌入学习和稀疏回归的无监督特征选择机制。现有特征选择方法往往把嵌入学习和特征排序孤立开来，而本项目提出统一框架，提出方法同时进行嵌入学习和稀疏回归学习，能够最大程度地利用嵌入学习和回归学习的优点。

成果综合统计
成果类型
数量
  • 期刊论文
  • 会议论文
  • 专利
  • 获奖
  • 著作
  • 58
  • 34
  • 0
  • 0
  • 0
期刊论文
Global structure constrained local shape prior estimation for medical image segmentation
Vehicle Detection and Motion Analysis in Low-Altitude Airborne Video Under Urban Environment
Constrained Nonnegative Matrix Factorization for Image Representation
Initialization Independent Clustering With Actively Self-Training Method
A probabilistic model for image representation via multiple patterns
Rapid pedestrian detection in unseen scenes
Lossless Data Embedding Using Generalized Statistical Quantity Histogram
Multivariate Multilinear Regression
A Nonlinear Adaptive Level Set for Image Segmentation
Efficient Search and Localization of Human Actions in Video Databases
Spatio-Temporal Laplacian Pyramid Coding for Action Recognition
Object-aware Power Line Detection using Color and Near-infrared Image
Spectral Embedded Hashing for Scalable Image Retrieval
Rank Preserving Discriminant Analysis for Human Behavior Recognition on Wireless Sensor Networks
Joint Embedding Learning and Sparse Regression: A Framework for Unsupervised Feature Selection
Universal Blind Image Quality Assessment Metrics Via Natural Scene Statistics and Multiple Kernel Le
Exploiting Local Coherent Patterns for Unsupervised Feature Ranking
Single-image Super-resolution via Local Learning
Travelogue Enriching and Scenic Spot Overview Based on Textual and Visual Topic Models
Joint Learning for Single-Image Super-Resolution via a Coupled Constraint
Nonrigid Structure-From-Motion From 2-D Images Using Markov Chain Monte Carlo
Parallelized Evolutionary Learning for Detection of Biclusters in Gene Expression Data
Intrinsic Image Decomposition Using Optimization and User Scribbles
Fast and Accurate Matrix Completion via Truncated Nuclear Norm Regularization
Visual saliency detection using information divergence
Tracking vehicles as groups in airborne videos
Depth-Aware Image Seam Carving
Embedding Motion and Structure Features for Action Recognition
Joint Optimization Toward Effective and Efficient Image Search
Learning Discriminative Key Poses for Action Recognition
Person Re-Identification by Regularized Smoothing KISS Metric Learning
Transductive Face Sketch-Photo Synthesis
Rank Preserving Sparse Learning for Kinect Based Scene Classification
Spatial-Constraint Hyperspectral Image Classification
Multi-Sensor Centralized Fusion without Measurement Noise Covariance by Variational Bayesian Approxi
A NOVEL METRIC BASED ON MCA FOR IMAGE QUALITY
Mammographic mass segmentation: Embedding multiple features in vector-valued level set in ambiguous
Non-goal scene analysis for soccer video
Face Sketch-Photo Synthesis and Retrieval Using Sparse Representation
Image Super-Resolution With Sparse Neighbor Embedding
Selecting Key Poses on Manifold for Pairwise Action Recognition
Enhanced Biologically Inspired Model for Object Recognition
A Comprehensive Survey to Face Hallucination
Efficient HOG human detection
A Relay Level Set Method for Automatic Image Segmentation
Zernike-Moment-Based Image Super Resolution
Detection of Sudden Pedestrian Crossings for Driving Assistance Systems
Saliency Detection by Multiple-Instance Learning
Semisupervised Dimensionality Reduction and Classification Through Virtual Label Regression
Aurora image segmentation by combining patch and texture thresholding
Biview face recognition in the shape-texture domain
Optimized Graph-Based Segmentation for Breast Ultrasound
Incremental Pairwise Discriminant Analysis based Visual Tracking
会议论文
Collaborative Kalman Filters for Vehicle Tracking
A-Optimal Non-negative Projection for Image Representation
Color Fractural Structure Model for Reduced-Reference Colorful Image Quality Assessment
ESTIMATING PATIENT-SPECIFIC SHAPE PRIOR FOR MEDICAL IMAGE SEGMENTATION
A NOVEL ALTERNATIVE ALGORITHM FOR LIMITED ANGLE TOMOGRAPHY
Local Learning-Based Image Super-Resolution
Single-Image Super-Resolution via Sparse Coding Regression
Putting Poses on Manifold for Action Recognition
PUTTING IMAGES ON A MANIFOLD FOR ATLAS-BASED IMAGE SEGMENTATION
Universal NR-IQA Metrics Based on Local Dependency
Linear SVM Classification using Boosting HOG Features for Vehicle Detection in Low-altitude Airborne
Utilizing Homotopy for Single Image Superresolution
Image Denoising via Weight Regression
Local Adaptive Dictionary based Image Denoising
Single-Image Super-Resolution Based on Semi-Supervised Learning
KLT Feature Based Vehicle Detection and Tracking in Airborne Videos
Accelerating Vehicle Detection in Low-altitude Airborne Urban Video
Face Sketch-photo Synthesis based on Support Vector Regression
ROBUST COLOR CORRECTION IN STEREO VISION
Multi-scale Dictionary for Single Image Super-resolution
Matrix Completion by Truncated Nuclear Norm Regularization
Sparse Representation for Blind Image Quality Assessment
Local Structure Divergence Index for Image Quality Assessment
A Biological Inspired Features Based Saliency Map
Target-oriented Shape Modeling with Structure Constraint for Image Segmentation
Designing and Selecting Features for MR Image Segmentation
Medical Image Segmentation Using Descriptive Image Features
Image Denoising via Improved Sparse Coding
Local Semi-Supervised Regression for Single-Image Super-Resolution
Segmenting Images by Combining Selected Atlases on Manifold
Learning Shape Statistics for Hierarchical 3D Medical Image Segmentation
Single Image Super Resolution with High Resolution Dictionary
Multimodal Learning for Multi-label Image Classification
High Quality Intrinsic Images Using Optimization
相关项目
光学影像处理与模式识别
期刊论文 183 会议论文 25 获奖 20
基于图像处理技术的输电线路故障诊断
期刊论文 15 会议论文 3 专利 2
基于动态CT多期图像的早期肝癌CAD技术研究
期刊论文 17 会议论文 8 专利 1 著作 1
视频检索关键问题研究及其在敏感信息识别中的应用
期刊论文 3 会议论文 4
基于图像识别的糖尿病视网膜病变自动诊断技术研究
期刊论文 10 会议论文 15 专利 8
基于机器视觉的新疆库尔勒香梨品质无损检测方法研究
期刊论文 5 会议论文 4
X射线数字图像智能识别及自动检测技术基础研究
期刊论文 8 会议论文 3
迁移学习及其在气象雷达数据分类中的应用
期刊论文 12 会议论文 7
基于图像处理的带钢表面缺陷自动识别技术
期刊论文 13 会议论文 1
李学龙的项目
光学影像处理与模式识别
期刊论文 183 会议论文 25 获奖 20