UN-SPIDER's Regional Support Office in Colombia, the Agustin Codazzi Geographic Institute (IGAC), is currently researching automatic flooded area extraction. The experts are implementing the Adaboost algorithm, which is a change detection procedure. This algorithm consists in creating a strong classifier that is built from a small group of weak classifiers. IGAC's approach uses the medium, mean and variance filters, each of these calculated using three different windows sizes: 3x3, 5x5, and 7x7. Then the experts calculate the difference-image between the filtered after-flood image and the filtered before-flood image.
The algorithm consists of two parts: The training and the application. In the first part, the algorithm is trained to create a model using the two differences-images and a ground truth image. For the application part, the experts used the differences-images and the model generated in the training part. Although some flooded parts were not detected, the algorithm did detect notable flooded regions.
