A Review of Full Resolution Quality Assessment for Multispectral Pansharpening

In recent years, multispectral pansharpening technology has developed rapidly aiming to integrate the spatial details of high-resolution panchromatic (PAN) images and the spectral information of low-resolution multispectral (MS) images to generate high-resolution MS images. The quality assessment of fused products has been a debated topic about pansharpening due to the lack of reference image. According to Wald's protocol, the quality assessment can be divided into reduced resolution (RR) validation and full resolution (FR) validation. RR validation is based on the synthesis of Wald's protocol which requires reference images and involves image degradation. The lack of higher resolution reference images and the influence of degraded filters make the FR quality evaluation become a research hotspot. This paper reviews the development of FR quality assessment for pansharpening by using the commonly known FR quality evaluation indicators to analysis 24 pansharpened methods. Finally, we find that FR quality assessment still has great challenges in accuracy, compared with RR quality assessment. FQNR and RQNR indexes have excellent performance in FR quality assessment. The pansharpening methods based on variational optimization-based (VO) methods and deep learning-based (DL) methods are controversial in numerical and visual evaluation.