斜模式高光谱成像的超分辨率重建方法研究

项目名称： 斜模式高光谱成像的超分辨率重建方法研究

项目编号： No.61501008

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 禹晶

作者单位： 北京工业大学

项目金额： 22万元

中文摘要： 高光谱成像过程中光谱分辨率与空间分辨率相互制约，光谱分辨率的增加导致空间分辨率的下降。本申请拟针对高光谱图像的空间分辨率提升问题，从斜模式成像和超分辨率重建两个方面，研究一种软硬件结合的斜模式超分辨率方法。内容包括：1)构建统一的斜模式成像模型，探讨不同倾斜角度和积分时间下的最优斜模式成像模型，设计斜模式超分辨率重建的技术框架；2)分析斜模式采样图像的频谱特性，估计成像系统的调制传递函数，研究斜模式采样图像的频谱去混叠方法；3)研究基于稀疏表示与结构自相似性的单幅图像超分辨率方法，分析高光谱图像的光谱特性和波段间的相关性，从数学模型上解决高光谱图像超分辨率重建中易发生的光谱失真问题；4)研究将斜模式采样特性融入稀疏表示超分辨率重建框架的方法，研究斜模式高光谱图像的超分辨率重建方法。本研究属于目前国际敏感性前沿技术之一，它的实施将为解决光谱分辨率与空间分辨率的相互制约提供一种可靠的技术手段。

中文关键词： 稀疏建模；图像盲复原；去模糊；超分辨复原；非局部正则化

英文摘要： During the acquisition of hyperspectral images, the spectral resolution and the spatial resolution restrict each other, and the spatial resolution usually decreases as the spectral resolution increases. In this project, we study a novel tilting-mode-based super resolution reconstruction technique to improve the spatial resolution of hyperspectral images from two aspects of tilting-mode sampling and super resolution reconstruction. The major issues of this project are as follows: 1) Construct a unified tilting-mode sampling model, explore the optimal tilting-mode sampling model under different tilting angles and integration time, and propose a general algorithmic framework for super resolution reconstruction; 2) Analyze the frequency characteristics of tilting-mode sampled images, estimate the modulation transfer function of the tilting-mode imaging system, and study how to remove aliasing for tilting-mode sampled images; 3) Study the structural self-similarity-based single image super resolution reconstruction method, and analyze the spectral characteristics and the correlation between bands of hyperspectral images and solve the super resolution reconstruction problem of hyperspectral images free of spectral distortion commonly occurring during reconstruction; 4) Discuss how to integrate the tiltling-mode sampling characteristics into the sparse representation-based super resolution reconstruction framework, and propose a sparse representation and structural self-similarity-based super-resolution reconstruction method for the tilting-mode sampled hyperspectral images. Our research belongs to the international research frontiers. A novel technological way will be provided in this project to solve the restriction problem bewteen the spectral resolution and the spatial resolution of hyperspectral images.

英文关键词： Sparse representation;blind image restoration;image deblurring;super resolution reconstruction;nonlocal regularization