项目名称: 内蕴特征空间基于低秩和稀疏分析的医学图像集处理方法
项目编号: No.61300067
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 自动化技术、计算机技术
项目作者: 李帅
作者单位: 北京航空航天大学
项目金额: 23万元
中文摘要: 特定器官的医学图像集所包含的信息大都集中在低秩、稀疏的内蕴结构中,但是如何自动提取该类结构并将其用于具有自适应和自学习特点的医学图像处理算法设计,还面临着诸多挑战。因此,本项目以稀疏表示和矩阵完备性分析理论为主要的数学工具,重点研究基于矩阵 "低秩+稀疏"表示理论的医学图像集处理方法。通过将基于各向异性热扩散理论的流形微分分析方法推广用于图像内蕴结构特征的度量和具有不变性的多尺度表示,可将图像在颜色空间的非线性相关性转化为特征空间的线性相关性。在此基础上,在内蕴特征空间对图像集处理问题进行基于"低秩+稀疏"表示的优化建模,可提高"低秩+稀疏"优化分析理论在非线性相关性分析和多尺度分析方面的处理能力,这对器官结构的自适应标注、基于图谱的器官分割以及图像集显著图的共生提取(co-generation)等应用具有重要的实用价值。目前,该领域的研究在国际上刚刚兴起,本项目有望取得重要创新成果。
中文关键词: 稀疏表示;低秩分析;图像内蕴特征;医学图像;图像集处理
英文摘要: The specific medical imageset usually concentrates on low-rank, sparse intrinsic structures. However, it remains chanllenging to extract and utilize such structures for the imageset processing method design that has self-adapting and self-learning charecteristics. Taking sparse representation and matrix completeness analysis theory as background mathematical tool, we focus on the low-rank and sparse analysis based medical imageset processing method. By extending the anisotropic heat diffusion based manifold differential analysis method to image for the measurement and invariant description of multi-scale intrinsic medical image structures, we can effectively transform the nonlinearity of image color space into linearity of intrinsic feature space. On that basis, we conduct low-rank and sparse representation based optimal modeling for some medical imageset processing requirements, which facilitates to the capalicity improvement of traditional low-rank and sparse representation theory in the aspects of nonlinear and multi-scale dataset analysis. Therefore, it has significant values in the adaptive labeling of organ anatomical structures,organ atlas based medical imageset co-segmentation,saliency map co-generation of medical imageset, etc. Specially, since the research in such fields is just a new rising topic in
英文关键词: sparse representation;low-rank analysis;image intrisinc feature;medical image;image set processing