Variational model-based very high spatial resolution remote sensing image fusion
Abstract
A remote sensing image fusion technique provides a mechanism for integrating multiple remotely sensed images to form an innovative image by using a certain algorithm for improving the spatial quality of the source image with minimal spectral distortion. Many algorithms, known as pan-sharpening algorithms, have been developed to improve the spatial resolution of multispectral (MS) images with a panchromatic (Pan) image. In the standard fusion methods, high spectral quality implies low spatial quality and vice versa. The utility of one Pan-sharpening model based on the variational model (VM) that consists of several energy terms is tested on very high spatial resolution images. In this model, the geometric structure matching term is used to inject the geometric structure of the Pan image, and the spectral matching term is utilized for preserving the spectral information. To balance the tradeoff between injecting the spatial information and preserving the spectral information, a static and a dynamic weight paradigm were introduced in this paper to control their relative contributions (static weights VM and dynamic weights VM). The evaluation of the experimental results on the QuickBird and WorldView-2 datasets shows that VM-based fusion models are better than the principal component analysis, Brovey transform fusion model, and Wavelet fusion model, and the dynamic weights VM performs better than the static weights VM. VM-based fusion models could be good options for very high spatial resolution remote sensing image fusion.