项目名称: 多重集典型相关分析的特征抽取理论及扩展研究
项目编号: No.61273251
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 自动化技术、计算机技术
项目作者: 孙权森
作者单位: 南京理工大学
项目金额: 80万元
中文摘要: 特征抽取是模式识别的基本问题,也是模式识别领域的研究热点。本项目以多表示数据的特征抽取与融合为研究对象,以多重集典型相关分析、分数维及稀疏表示的前沿思想为出发点,集中解决与实际应用密切相关的多组特征抽取理论及其相关技术,探索有关前沿理论在多重集典型相关分析中的合理推广与拓展,从而突破其中的关键理论与技术,构建一套行之有效的多组特征抽取与融合的理论框架。本项目拟开展如下研究工作:(1)引入监督信息、核技术以及局部化思想,深化研究多重集典型相关分析的理论及多组特征抽取技术;(2)基于分数维的思想,开展分数维(多重集)典型相关分析理论框架的构建与应用研究;(3)以多组字典的共同学习和特征的多稀疏表示为主要研究内容,深入开展稀疏多重集典型相关分析的理论研究。本项目对于推动多表示数据的特征抽取与融合技术的研究具有重要的理论与实际意义。
中文关键词: 特征抽取;多重集典型相关分析;特征融合;分数维;稀疏表示
英文摘要: Feature extraction is a fundamental problem and has become a research focus in pattern recognition. In order to solve the critical techniques in multiset feature extraction, the project first investigates in depth multiset canonical correlation analysis with applications to feature extraction and fusion in multi-representation data. Second, the project studies the generalization and extensions of multiset canonical correlation analysis based on fractal dimension space and sparse representation. Finally, we will break through the related crucial theory and techniques to construct a set of effective and robust multiset feature extraction and fusion methods. The project will perform the following research work: (1) Introducing supervised information, kernel learning and locally linear methods respectively to multiset canonical correlation analysis, build multiset feature extraction techniques in multi-representation data; (2) Based on the idea of fractal dimensions, construct a fractal-dimension (multiset) canonical correlation analysis framework; (3) Based on the learning of multiple dictionaries simultaneously and the united sparse representation of multiple features, we are intended to intensively study the theory research of sparse multiset canonical correlation analysis. The research in the project will vastly
英文关键词: Feature extraction;Multiset canonical correlation analysis;Feature fusion;Fractal dimension;Sparse representation