非负矩阵分解和Curvelet在遥感图像融合中的应用  

Application of Non-negative Matrix Factorization and Curvelet to Remote Sensing Image Fusion

作　　者：曾立庆[1,2] 童怀水[2] 

机构地区：[1]东华理工大学江西省数字国土重点实验室,江西抚州344000 [2]东华理工大学理学院,江西抚州344000

出　　处：《东华理工大学学报（自然科学版）》2013年第4期415-418,共4页Journal of East China University of Technology(Natural Science)

基　　金：江西省数字国土重点实验室开放研究基金项目(DLLJ201013)

摘　　要：非负矩阵分解是一种提取图像原始信息局部特征的新方法,第二代Curvelet变换是一种效果较好的多尺度变换分析方法。结合两者特征提出一种基于NMF和Curvelet的遥感图像的融合方法,首先对已配准的多光谱图像和全色图像进行Curevelet分解,得到各层系数(Coarse、Detail和Fine尺度层)。然后对Coarse尺度层(低频系数)进行NMF分解,提取出包含特征基的低频系数;对Detail和Fine尺度层(高频系数)采用方差为测度参数进行邻域融合。最后进行Curevelet逆变换得到融合图像。实验结果表明,该方法的融合图像能较好地保留光谱信息,并在空间细节信息上得到改善,优于小波方法、Curvelet等方法。Nonnegative matrix factorization is a new method for extracting local features from original information of images. The second generation Curvelet is a preferable multiscale analysis algorithm. A new method of remote sensing image was proposed based on NMF and Curvelet combining with both characteristics. Firstly, The regis tered images of Multispectral and Panchromatic were decomposed using curvelet to get coefficients of each layer （Coarse,Detail and Fine scale layer）. Secondly, NMF was applied to coarse scale layer（ low frequency coeffi cients）, and the resultant feature base was just the fusion result of low frequency suhhand coefficients ; the selec tion principle of detail and fine scale layer（ high frequency coefficients） was the neighboring region variance maxi mum. Finally, the fused image was obtained by performing the inverse curvelet on the combined coefficients. The experimental results show that the proposed methods can effectively keep spectrum information, and improve in the detail information of space. And it outperforms waveletbased and curveletbased fusing algorithms.

关 键 词：非负矩阵分解 CURVELET变换 遥感图像 图像融合 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]