Evaluation of Sun Glint Correction Algorithms for High-Spatial Resolution Hyperspectral Imagery

Abstract

Sun glint correction algorithms were tested on a hyperspectral image containing cross-track sun glint. Spatial profiles of pixel radiance by pixel position were compared and slope values were calculated. The algorithms of Hedley et al., Lyzenga et al., and Joyce over-corrected for sun glint in the visible and near-infrared wavelengths. The method proposed by Kuster et al. was the weakest performer during visual comparison with the other method results. Spectral plots of corrected spectra to the original spectra were compared. Comparisons were performed on spectra from pixels with low and high amounts of sun glint. Spectra were compared within the sun glint corrected images and between the corrected images and the original image. Correlation values were calculated for each spectral comparison and averaged for each sun glint correction algorithm. The Lyzenga et al. sun glint correction algorithm had the highest average correlation value of 0.977 and is recommended for reducing sun glint in hyperspectral imagery when spectral integrity is required.