Pansharpening Based on Spectral-Spatial Dependence for Multibands Remote Sensing Images

Abstract

Pansharpening based on detail injection (DI) explores panchromatic (PAN) images and injects their spatial geometrical information into multispectral (MS) images. Traditional DI algorithms fuse remote sensing images without considering spectral and spatial dependence, which causes spectral distortion. To overcome this problem, this article proposes an advanced version of the DI model that not only achieves adaptive DI but also addresses spectral-spatial dependence. This proposed method is called pansharpening based on spectral-spatial dependence for multi-bands remote sensing images. In the proposed method, three parameters including an adaptive DI based on the correlation between MS and PAN images, the spectral dependence based on the MS band and pixel dependence, and the spatial dependence based on the detail offset amplitude are designed. An improved GDI model based on DI and the combination of the spectral and spatial dependence (SSD) is constructed. In the model, SSD enables the fused image to retain the spectral fidelity, and DI ensures the sharpness of the fused image. A performance test is conducted on various satellite datasets and the results are compared with those of several state-of-theart fusion methods. The results show that the proposed method benefits pansharpening.

Bayesian decision based fusion algorithm for remote sensing images

“…It is well-known that a MS image can be decomposed into intensity (I), hue (H), saturation (S) by IHS transformation 18 . The purpose of this study is to design a novel fusion rule to integrate the PAN image and the I component of the MS image to obtain a new intensity component I new , and replace I with I new to obtain a desired HRMS image.…”