项目名称: 低秩矩阵恢复算法及其在图像处理中的应用
项目编号: No.11271367
项目类型: 面上项目
立项/批准年度: 2013
项目学科: 数理科学和化学
项目作者: 王来生
作者单位: 中国农业大学
项目金额: 65万元
中文摘要: 低秩矩阵恢复和张量恢复是优化领域和信息科学领域最近研究的热点,在推荐系统、图像处理和计算机视觉等方面已经找到重要的应用。现有的算法计算量大、速度慢、对于大规模问题效果不好,这使得它在很多场合不能充分发挥其作用和优势。本项目主要研究低秩矩阵恢复以及张量恢复问题的理论、算法和它在图像处理中的应用。主要包括:从理论上进一步研究在新的测量算子作用下恢复低秩矩阵的RIP条件以及Mp非凸松弛模型RIP条件的改进;针对非凸的Mp极小化模型来设计有效的算法,使得对于大规模矩阵能够得到良好的效果;应用Mp范数极小化的算法在图像背景的特征提取过程中尽量保存完整的边缘信息;研究基于矩阵恢复的低秩矩阵近似分解算法,并将其结合支持向量机应用到核空间的特征提取中;从理论和算法两方面利用已有的研究成果对张量恢复问题进行研究,并应用张量恢复的方法在多渠道图像和视频方面进行图像修复使之得到更好的试验效果。
中文关键词: 矩阵恢复;低秩矩阵;张量恢复;图像处理;
英文摘要: Low rank matrix completion and tensor completion have recently become popular in optimization and information sciences. Their applications can now also be found in recommendation systems, image processing and computer vision. The existing algorithms in many cases cannot sufficiently play their role and advantage because of its large amount of calculation, slow speed and the bad effect for large-scale matrix. In this project, we mainly study the theory, algorithm and application in image processing of Low Rank Matrix Completion and Tensor Completion. Our research content as follows: Firstly, in theory we will further research that under which restricted isometry property (RIP) condition on the new linear transformation we can obtain the exact low-rank matrix solution by solving the nuclear norm minimization and that how to improve RIP conditions for exact low rank matrix recovery via nonconvex relaxations Mp-minimization. Secondly, we will propose the effective algorithms for the nonconvex relaxations Mp-minimization to solve efficiently the problem which the size of the matrix is large. Thirdly, we will try to maintain integrity of edge information in the process of the feature extraction by the algorithms of Mp-minimization. Forthly, we study low-rank decomposition of kernel matrix based on matrix completion, a
英文关键词: matrix completion;low rank matrix;tensor completion;image processing;