基于L1范数优化模型的遥感图像条纹去除方法  

The method based on L1 norm optimization model for stripe noise removal of remote sensing image

作　　者：李凯 李文力[1,2,3] 韩昌佩 LI Kai;LI Wen-Li;HAN Chang-Pei(Shanghai Institute of Technical Physics,Chinese Academy of Sciences,Shanghai 200083,China;Key Laboratory of Infrared Detection and Imaging Technology,Chinese Academy of Sciences,Shanghai 200083,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院上海技术物理研究所,上海200083 [2]中国科学院红外探测与成像技术重点实验室,上海200083 [3]中国科学院大学,北京100049

出　　处：《红外与毫米波学报》2021年第2期272-283,共12页Journal of Infrared and Millimeter Waves

基　　金：中国科学院上海技术物理研究所创新专项(CX-208)。

摘　　要：从条纹噪声的结构属性进行分析,通过分离出条纹成分来实现去条纹的目的。在优化模型中,基于L1范数的正则化表示条纹的全局稀疏特性;基于差分的约束条件用于描述条纹方向上的平滑度和条纹垂直方向上的不连续性。为了更好地保护图像的细节信息,在条纹垂直方向的约束上引入了边缘权重因子,最后通过交替方向乘子法(ADMM)对所提模型进行求解和优化。用多通道扫描辐射计(AGRI)获取的在轨数据对算法进行了验证并与典型方法进行了比较,结果表明,消除条纹噪声的同时更好地保留了细节信息,并且呈现出较好的定性和定量结果。Structural properties of stripe noise are analyzed and the purpose of destriping is achieved by separating the stripe components.In the proposed optimization model,the L1-norm-based is used to describe global sparse property of stripes.In addition,difference-based constraints are adopted to describe the smoothness and discontinuity in the alongstripe and across-stripe directions,respectively.In order to better protect the detailed information of an image,an edge weighting factor is introduced in the constraints of across-stripe direction.Finally,the proposed model is solved and optimized by the alternating direction method of multipliers(ADMM).The algorithm is verified by the in-orbit images obtained by Advanced Geosynchronous Radiation Imager(AGRI)in comparison with typical destriping methods.Experimental results show that the proposed algorithm completely eliminates the stripe noise and preserves more details,which shows better qualitative and quantitative result.

关 键 词：L1稀疏优化模型 图像去条纹 边缘权重因子 交替方向乘子法 AGRI图像 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]