Accurate Shadow Detection From High-Resolution Satellite Images

High-resolution satellite images contain a huge amount of information. Shadows in such images generate real problems in classifying and extracting the required information. Although signals recorded in shadow area are weak, it is still possible to recover them. Significant work is already done in shadow detection direction but, classifying shadow pixels from vegetation pixels correctly is still an issue as dark vegetation areas are still misclassified as shadow in some cases. In this letter, a new image index is developed for shadow detection employing multiple bands. Shadow pixels are classified from the index histogram by an automatic threshold identification procedure. The whole approach is applied on different study areas and high accuracies are achieved (average of 97%). The linear correlation method is then applied to compensate the classified shadow pixels. Two standard approaches of shadow detection are then applied to the same study areas to validate the proposed approach. The results show that the proposed approach achieves the best results. It also gives robust shadow detection results in classifying shadow from vegetation pixels comparable to the other two considered standard approaches.