项目名称: 基于机器学习的局部图像特征描述与融合机制研究
项目编号: No.61203277
项目类型: 青年科学基金项目
立项/批准年度: 2013
项目学科: 自动化学科
项目作者: 樊彬
作者单位: 中国科学院自动化研究所
项目金额: 24万元
中文摘要: 局部图像特征描述与匹配是计算机视觉研究中的一个基本问题,影响着许多相关应用算法的发展,例如物体识别、三维重建和图像检索。本项目针对传统的局部图像特征描述子设计方法上的不足,旨在通过机器学习方法,深入研究数据驱动型的局部图像特征描述子的自动设计方法,同时研究多种特征描述子的融合机制,提出新的特征描述与匹配方法。主要研究内容包括:(1)从"特征描述=底层特征提取+底层特征融汇"的观点出发,分别研究基于机器学习的底层特征提取和底层特征融汇方法;(2)从数据降维和特征选择的观点出发,研究基于不变性度量的图像至向量的展开方式以及用于特征匹配的子空间学习和特征选择方法;(3)基于机器学习,分别研究特征层和决策层的多特征描述子融合机制。本项目的研究成果可望丰富特征描述、特征匹配、机器学习等相关领域的研究内容,为图像特征描述和匹配提供新的研究思路和方法。
中文关键词: 特征描述;特征匹配;;;
英文摘要: Local image descriptors construction and matching is a fundamental problem in computer vision, and has a profound influence on the related applications, such as object recognition, 3D reconstruction and image retrieval. To alleviate the shortcomings of the traditional methods for local descriptor construction, this proposal aims to propose data-driven methods for constructing local descriptor based on machine learning. Meanwhile, it also studies multiple descriptors fusion strategies to further improve their performance. The key issues include: (1) Modeling the local image description as a process of low-level feature extraction followed by feature pooling, then focused on the research of low-level feature extraction and feature pooling methods through machine learning;(2) Modeling the local image description as a process of dimension reduction or feature selection, then research on the methods for robustly converting images into vectors based on invariants, and research on the subspace learning as well as feature section methods for feature matching;(3) Research on the application of machine learning for multiple descriptors fusion in feature-level and in decision-level respectively. The outcomes would enrich methodology and provide novel ideas for feature description, feature matching, feature fusion and machi
英文关键词: Feature Description;Feature Matching;;;