The effects of segmentation-based shadow removal on across-date settlement type classification of panchromatic QuickBird images

摘要

Settlement classifiers for multitemporal satelliteimage analysis have to overcome numerous difficulties relatedto across-date variances in viewing- and illumination geometry.Shadow anisotropy is a prominent contributing factor in classifierinaccuracy, so methods are introduced in this study to enableminimum-supervision classifier design that mitigate the effectsof shadow profile differences. A segmentation-based shadow detectoris proposed that utilizes a panchromatic segment mergingalgorithm with parameters that are robust against dynamic rangevariances seen in multitemporal imagery. The proposed shadowdetector improves on the settlement classification accuracyachieved by fixed threshold detection paired with shadow removalin the presented case-study. The relationship between shadowdetection accuracy and settlement classification accuracy is investigated,and it is shown that shadow removal produces greatersettlement accuracy improvements for across-date experimentsspecifically.