Development of a robust haze removal algorithm: assessment using temporally invariant targets

A robust haze-removal algorithm based on the Haze Optimized Transformation (HOT) is applied to ETM+ imagery. Its effectiveness is evaluated using temporally invariant targets identified in a multi-temporal data set of scenes of WRS-2 Path15/Row29. In the data set, clear scenes serve as ´benchmarks´ against which the effectiveness of haze removal can be assessed. In the quantitative assessment, two hundred coniferous forest patches were randomly sampled and served as ´benchmark´ targets since they exhibit both spatial and temporal radiometric invariance on clear data. Comparisons were made of the radiometric characteristics of the patch population in hazy vs. de-hazed and de-hazed vs. clear image pairs. Results show that image contaminated by low to moderate haze can be significantly improved. Reductions of 20 to 90% in intra-scene radiometric dispersion have been achieved with the de-hazed scenes exhibiting residual dispersion levels comparable to those of the benchmark scenes.