Hyperspectral and Multispectral Image Fusion Using 3D Wavelet Transforms

Abstract

The fusion of multispectral (MSI) and hyperspectral (HSI) images is a crucial technique in various fields such as remote sensing, medical imaging, and agricultural monitoring. MSI captures light across several specific spectral bands, while HSI provides detailed spectral information across contiguous bands. Combining these two types of images leverages the high spatial resolution of MSI and the rich spectral content of HSI, creating a single, high-resolution image that is both spatially and spectrally informative. Traditional wavelet based fusion methods often employ a single wavelet across all dimensions, which can result in suboptimal outcomes due to the different characteristics of spatial and spectral data. This paper explores the use of 3D wavelet transforms with varied wavelets across dimensions to improve the fusion process. Experiments conducted on the ICASSP HyperSkin Challenge dataset showed that a combination of Daubechies on the spatial dimensions and Coiflets on the spectral dimension obtained higher fidelity and SSIM when compared to simpler fusion methods.