Quantitative assessment of a haze suppression methodology for satellite imagery: effect on land cover classification performance

Abstract

A number of haze removal methods for satellite imagery have been described in the literature, but few papers have quantified their effectiveness within the context of a postprocessing applications scenario. The haze optimized transform (HOT) approach for visible-band correction is described, and its impact on classification performance is evaluated. Assessment is conducted at three levels: radiometric level of accuracy, improvement in interclass separability, and classification accuracy. Results obtained from hazy scenes and their dehazed counterparts are compared with those from reference or "benchmark" clear scenes. Radiometric analyses of pseudoinvariant features (dense forest stands) indicate that effective haze reduction can be realized for differential HOT response levels of up to 20 for Landsat Thematic Mapper scenes. This level of atmospheric contamination is severe enough to result in significant thematic class confusion, wherein visible-band radiances of vegetated areas are at levels normally associated with urban features under clear sky conditions.