Framework for feature extraction of natural objects

Feature extraction of natural objects is still a very difficult task with todays image processing techniques. High resolution aerial or satellite data can be acquired with reasonable effort, but accuracy and correctness of available extraction methods are lagging behind. Instead of propagating the ultimate segmentation algorithm, a hypothesis-based approach is presented, which tries to combine established and new image processing methods with each other. Similar to the varimax criteria of principal component analysis, extraction methods should be as independent as possible from each other.