IKONOS图像的线性回归波段拟合融合方法

Band simulation based pan-sharpening algorithm by linear regression for IKONOS imagery

讨论基于线性回归波段拟合的空间细节信息提取方法的可行性。首先通过全色与多光谱图像构造线性回归方程,根据全色图像的高频成分设置最小二乘求解的权系数,然后利用回归系数构造低分辨率的全色图像,提取空间细节信息,最后将空间细节注入多光谱图像中进行融合。通过IKONOS全色和多光谱图像的融合实验,比较了本文方法与基于光谱响应函数的方法,结果表明:采用本文方法提取的空间细节信息进行融合,能达到甚至超过基于光谱响应函数方法的融合质量;相对于与FastIHS融合方法,本文方法的融合质量也有较大的提高。

In the general fusion framework of optical remote sensing image fusion, the extraction of spatial detail information is one of the two key aspects for the fusion quality. This paper discusses the possibility of constructing the low resolution panchromatic image based on band simulation using linear regression. Firstly, a linear regression equation between panchromatic and multispectral images was modeled, followed by setting weight metric from the corresponding high frequency component of the panchromatic image; then, low resolution panchromatic image was simulated by the regression parameters got in the first step and the spatial detail was extracted; finally, the pan-sharpening procedure was implemented with the spatial detail on IKONOS multispectral and panchromatic imagery. Compared with the method of achieving low resolution panchromatic image based on spectral response function, the proposed method can perform as good as, and sometimes even better than that according to four experiments. Also, the proposal shows its superiority over fast intensity-hue-saturation method.