空谱联合相关性驱动的高光谱图像概率图修补模型与算法

项目名称： 空谱联合相关性驱动的高光谱图像概率图修补模型与算法

项目编号： No.61301215

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 无线电电子学、电信技术

项目作者： 刘红毅

作者单位： 南京理工大学

项目金额： 25万元

中文摘要： 高空中水汽、云层等因素会引起高光谱图像数据中的部分光谱和空间信息丢失，补全这些丢失信息对后续的高光谱图像处理具有非常重要的理论和应用价值。本项目以高光谱图像修补为具体科学问题，联合光谱曲线相似性度量，深入分析高光谱图像patch局部与非局部相关性表征机理和先验度量；基于局部和非局部邻域系统的簇和势函数，构建具有不同拓扑结构的概率图模型；在最大后验概率框架下，提出联合空谱相关的高光谱图像概率图修补模型，并提出高效的并行优化算法。本项目对于推动高光谱图像去噪、分类、识别等后续研究具有重要的理论意义。同时，本项目所提供的方法和技术在地质勘查，植被遥感、大气遥感等诸多领域具有广阔的应用前景。

中文关键词： 高光谱图像；马尔科夫随机场；图像修补；正则化；低秩表示

英文摘要： Due to the aerological water vapor, clouds and cloud shadows, part of spectral and spatial information of hyperspectral image may be missing, therefore inpainting is an essential task for the post-processing of hyperspectral image processing. In this project, we make a study on hyperspectral image inpainting by fully exploiting spatial-spectral joint correlation priors of hyperspectral patch in a local and nonlocal ways . Based on the cliques and potential functions which are defined by the spatial neighborhoods and multispectral bands, an undirected probabilistic graph with different topology structure is constructed. Finally, a patch based hyperspectral inpainting model via probabilistic graphical model is proposed under the framework of MAP, and accelerated using an efficient parallel optimization algorithm. The research of the project will enrich and promote the development of hyperspectral inpainting theory and algorithm, and has a great significance to hyperspectral denoising, classification and identification. Meanwhile, the project has a very important theoretical value as well as a very broad application prospect to hyperspectral image processing techniques, geological exploration, remote sensing of vegetation and atmospheric remote sensing,etc.

英文关键词： Hyperspectral image；Markov random field；Image inpainting；Regularization；Low-rank representation