Haze removal based on advanced haze-optimized transformation (AHOT) for multispectral imagery

Abstract

Ever-present spatially varying haze contamination in satellite scenes limits applications using visible and near-infrared bands of low-temporal-resolution multispectral satellite images. A relative atmospheric correction technique, the virtual cloud point (VCP) method, which is based on advanced haze-optimized transformation (AHOT), is developed for haze removal. It is an improved algorithm of the previous dark-object subtraction (DOS) based on haze-optimized transformation (HOT). In AHOT, extra steps are added to HOT to remove confusion caused by some land-cover types. The VCP method uses the upper bound of the histogram, as well as the lower bound, so that it enlarges the digital number (DN) variance reduced by haze, which is not considered in DOS. To evaluate this algorithm, hazy subsets of one Landsat Thematic Mapper (TM) and one QuickBird image are employed. Through before-and-after comparison using true-colour images and the normalized difference vegetation index (NDVI), it proves that the VCP method based on AHOT is apparently better than DOS based on HOT, when haze is distributed over urban areas where vegetation is sparse.