A theoretical and practical survey of image fusion methods for multispectral pansharpening

Pansharpening fuses the spatial features of a high-resolution panchromatic (PAN) image with the spectral features of a lower-resolution multispectral (MS) image to generate a spatially enriched MS image. Numerous pansharpening strategies have been developed for more than three decades, which forces the analysts who intend to apply pansharpening to choose from various pansharpening techniques. Hence, this study aims to investigate the performances of many conventional and state-of-the-art pansharpening techniques in order to guide the analysts in this regard. To this aim, the spectral and spatial structure fidelity of the pansharpened images produced from a total of 47 pansharpening methods were evaluated qualitatively and quantitatively. The methods examined were from six pansharpening methods categories, including Multiresolution Analysis (MRA)-based, Component Substitution (CS)-based, Colour-Based (CB), Deep Learning (DL)-based, Variational Optimization (VO)-based and hybrid techniques. The methods in the MRA, DL, CB and VO category were found to exhibit the best pansharpening performances; whereas the hybrid and CS-based techniques showed the poorest performances. We believe that the outcomes of this study will guide the analysts who are in the need to apply pansharpening for their applications.

Automated summary

The study investigated the performance of various conventional and state-of-the-art pansharpening techniques, finding that methods in the Multiresolution Analysis (MRA), Deep Learning (DL), Colour-Based (CB) and Variational Optimization (VO) categories exhibited the best pansharpening performances, while hybrid and Component Substitution (CS)-based techniques showed poorer performances.