基于稀疏表示的遥感图像融合方法  

Remote Sensing Image Fusion Based on Sparse Representation

作　　者：尹雯[1] 李元祥[1] 周则明[2] 刘世前[1] 

机构地区：[1]上海交通大学航空航天学院,上海200240 [2]解放军理工大学气象学院,江苏南京211101

出　　处：《光学学报》2013年第4期259-266,共8页Acta Optica Sinica

基　　金：国家自然科学基金(41174164;61174196)资助课题

摘　　要：为了提高融合后多光谱(MS)图像的质量,提出一种基于稀疏表示的遥感图像融合方法。建立MS图像与其亮度分量之间的线性回归模型;利用训练的高、低分辨率字典分别对全色图像和MS图像进行稀疏表示,并根据线性回归模型获得MS图像亮度分量稀疏表示系数;根据全色图像和亮度分量的稀疏表示系数提取细节成分,并在通用分量替换(GCOS)融合框架下注入到MS图像各波段的稀疏表示系数中;进行图像复原得到高空间分辨率的MS图像。由于稀疏表示可有效地刻画信号的内部结构与特征,融合后的MS图像能够在提高空间分辨率的同时,较好地保留原始MS信息。IKONOS MS图像的融合结果表明,该方法在光谱保持和空间分辨率提高方面优于其他传统的遥感图像融合方法。In order to improve multi-spectral （MS） image fusion quality, a new pan-sharpening method based on sparse representation is proposed. A linear regression model between the MS image and its intensity component is established. The sparse coefficients of both panchromatic image and MS image are obtained by two dictionaries which are trained to have the same sparse representations for each high-resolution and low-resolution image patch pair. The coefficient of intensity can also be obtained via the linear regression model and the coefficients of MS bands. Then, the sparse coefficients are fused in the general component substitution （GCOS） fusion framework. The fused sparse coefficients are used to reconstruct a high-resolution MS image. As the inherent characteristics and structure of signals are by via sparse representation more efficiently, the proposed method can preserve spectral and spatial details of the source images well. Experimental results on IKONOS satellite images demonstrate the superiority of the proposed method in both spatial resolution improvement and spectral information preservation.

关 键 词：遥感 图像融合 稀疏表示 通用分量替换融合框架 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]