基于多尺度金字塔的多光谱图像Pan-Sharpening方法

Pan-Sharpening Method for Multispectral Images Based on Multi-scale Pyramids

针对传统高分辨率全色(high-resolution panchromatic,HRP)图像融合中发生的频谱失真等问题,提出了一种多尺度金字塔方法来锐化低分辨率多光谱(low-resolution multi-spectral,LRM)图像.该方法利用HRP图像中的冗余补丁重建高分辨率多光谱(high-resolution multi-spectral,HRM)图像.首先,从低分辨RP图像创建金字塔;然后,通过利用该金字塔作中同一层以及较低层的每个补丁之间的关系,重构上采样的LRM频带;最后,从上采样的多光谱带估计高分辨率强度分量.利用不同层次的相似结构更详细地重构HRM的波段特性.在多尺度过程中利用自相似性,从可用HRP和LRM图像中构建HRM图像,减少了空间失真.此方法在强度分量的重建中使用底层HRP图像来减小HRP和强度分量之间的不相似性,从而减少频谱失真.实验结果表明,此方法能有效地保留源图像的光谱和空间信息,且性能优于其他方法.

Aiming at the problem of spectral distortion in traditional high-resolution panchromatic(HRP)image fusion,a multi-scale pyramid method has been proposed to sharpen low-resolution multi-spectral(LRM)images.In this method,the redundant patches in HRP images have been used to reconstruct high-resolution images.The high-resolution multi-spectral(HRM)image firstly creates a pyramid from the degraded HRP image.Then,by means of the relationship between each patch in the same layer and in the lower layer of the pyramid,the upsampled LRM band has been reconstructed and the high-resolution intensity component been estimated from the upsampled multi-spectral band.The band characteristics of HRM have been reconstructed in more details by means of similar structures of different layers,and the self-similarity been used to construct HRM images from available HRP and LRM images in multi-scale process,reducing the spatial distortion.In this method,the underlying HRP image has been used to reduce the dissimilarity between the HRP and the intensity component in the reconstruction of the intensity component,thus reducing the spectrum distortion.Experimental results show that the proposed method can effectively preserve spectral and spatial information of the source image,and its performance is better than other methods.