项目名称: 动态纹理建模与应用的张量方法研究
项目编号: No.11301137
项目类型: 青年科学基金项目
立项/批准年度: 2014
项目学科: 数理科学和化学
项目作者: 周丙寅
作者单位: 河北师范大学
项目金额: 22万元
中文摘要: 本项目主要研究动态纹理建模与应用的张量方法.动态纹理是人们认识事物的重要视觉信息,通常会产生巨大量的高维数据;张量是高维数据的自然表示形式,能够保持数据的内在结构,张量方法是高维数据处理和分析的潜在有力方法.以动态纹理合成、识别和分割以及建立在其上的视频分析应用为驱动,完善和发展张量特征值和张量分解的理论与方法,提出高维数据紧凑表示的概念并给出其严格定义和定量刻画方法,为张量分析和高维数据表示提供新的研究思路和方法.反过来,将这些理论研究成果应用于对动态纹理的深刻认识,从而更有效的解决动态纹理合成、识别和分割问题,以及建立在其上的视频分析关键问题,例如动态背景建模和视频内容刻画与分类等.
中文关键词: 张量分解;张量特征值;紧凑表示;动态纹理;视频分析
英文摘要: This project mainly studies the tensor method for dynamic texture modeling and applications. Dynamic textures are powerful visual cues for people to understand things, which usually generate an enormous size of high-dimensional data. Tensors are natural representations of high-dimensional data preserving their intrinsic structure, and tensor methods are promising methods to process and analyze high-dimensional data. Driving by the applications of dynamic texture synthesis, recognition, segmentation and related video analysis applications, we improve and develop the theories and methods of tensor eigenvalue and tensor decomposition, and propose the strict definition and quantitatively characterize method of compact representation for high-dimensional data. These will provide new ideas and methods for tensor analysis and data representation. In turn, by applying these theoretical results for understanding dynamic textures, the problems of dynamic texture synthesis, recognition, segmentation and the related key problems such as dynamic background modeling and video content classification of video analysis will be handled more effectively.
英文关键词: tensor decomposition;tensor eigenvalue;compact representation;dynamic textures;video analysis