ABSTRACT

In this paper, the characteristics of low-resolution multispectral image (LRMS) and panchromatic image (PAN) fusion methods are investigated. Recently, the sparse representation (SR) based methods and the injected details (ID) based methods have been combined as the pansharpening method based on SR of ID. This novel method can gain better results than the former ones, but it also faces two disadvantages, i.e., the choice of using the raw patches as dictionary and using the SR to all parts of the aiming image are not so optimal, which will cause inaccurate representation and sharper than it should be in the smooth area. Thus, we propose a new pansharpening method based on SR of classified ID over featured dictionary to learn dictionary from the featured details and fuse different patches diversely to overcome the drawbacks mentioned above and enhance the quality of aiming image. The experimental results using QuickBird and WorldView2 show that the proposed method can achieve remarkable spectral and spatial quality.

摘要

本文研究了低分辨率多光谱图像（LRMS）和全色图像（PAN）融合方法的特点。最近，基于稀疏表示 (SR) 的方法和基于注入细节 (ID) 的方法已结合作为基于 ID 的 SR 的全色锐化方法。这种新颖的方法可以获得比前一种更好的结果，但它也面临两个缺点，即使用原始补丁作为字典和对瞄准图像的所有部分使用SR的选择不是那么优化，这会导致不准确表示并且比平滑区域中应有的更清晰。因此，我们提出了一种新的基于特征字典上的分类 ID SR 的全色锐化方法，从特征细节中学习字典并不同地融合不同的补丁，以克服上述缺点并提高瞄准图像的质量。使用 QuickBird 和 WorldView2 的实验结果表明，所提出的方法可以获得显着的光谱和空间质量。