基于超完备稀疏分解的高光谱图像超分辨率复原技术研究

项目名称： 基于超完备稀疏分解的高光谱图像超分辨率复原技术研究

项目编号： No.61201361

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 电子学与信息系统

项目作者： 王素玉

作者单位： 北京工业大学

项目金额： 24万元

中文摘要： 本项申请针对实际应用中对高分辨率高光谱图像的广泛需求，开展基于超完备稀疏分解的高光谱图像超分辨率复原技术研究,首先研究一种基于地物类别的高光谱图像稀疏分解算法，通过学习训练的方式寻找各类典型地物中最具代表性的光谱特征构成冗余字典，将高光谱图像中的每个像元分解为相应原子的最优线性组合的形式。冗余字典的设计过程引入结构相似度指标SSIM以提高所设计的字典对于图像中纹理、结构信息的描述能力。进而以此为基础，研究一种基于双字典的高光谱图像超分辨率复原算法，通过约束学习构建一组高、低分辨率相对应的冗余字典对，使得相互对应的一组高、低分辨率样本能够以相同的稀疏表示系数实现稀疏分解。在超分辨率复原过程中，首先将低分辨率图像基于低分辨率冗余字典进行稀疏分解，然后利用该稀疏分解系数基于对应的高分辨率冗余字典，重建高分辨率的高光谱图像。通过基于降质模型的约束优化进一步提高所重建图像的保真度。

中文关键词： 冗余字典；超分辨率复原；空间分辨率；稀疏表示；光谱分辨率

英文摘要： To meet the wide requirements of high resolution hyperspectral images in all kinds applications, this project is planned to research on technologies of overcomplete sparse decomposition based hyperspectral image super-resolution restoration. A material based hyperspectral image sparse decomposition algorithm is first investigated, which is to find the most representive spectral characters to establish a sparse dictionary,and make all the pixels of the hyperspetral image can be decomposed into best linear combinations of these itoms. To further improve its ability to describe texture and structer informations of the hyperspectral images, the index of SSIM(structural similarity index) is introduced during the process of the dictionary training. Furthermore,a double dictionary based hyperspectral image super resolution scheme is investigated. A pair of high-low resolution sparse dictionary is established first,which is to ensure that a pair of high-low resolution image can be sparse decomposed by the same coefficients. During the process of super resolution,the low resolution image is first dcomposed according to the correspond low resolution dictionary.Then the high resolution image is reconstructed by use of the same coefficients and the correspond high resolution dictionary. A image model of the hyperspectral im

英文关键词： redundant dictionary；super resolution；spatial resolution；sparse representation；special resolution